So... yes and no.
First, the "No"
Technically most of the other answers saying "no" are correct, if incomplete. In the absolute strictest sense, they are not obligated to provide data dumps ever again, at least not by the license terms¹. That's because it doesn't require sharing anything we submit - the license gives them the right to stop using licensed content at any time. They also aren't specifically required by the license to collate or upload a full dump, much less to do so in some particular format.
Now the (qualified) "Yes".
A commonly missed or misunderstood part of the CC-BY-SA license is the "ShareAlike" provision. SE picked this license "flavor" because the SA version allows them to "remix" submissions freely, using them pretty much any way they want. But this freedom comes with a catch: they have to give others the same flexibility in turn.
Even derivative, or "adapted materials" must be shared under the same CC-BY-SA license. In fact, by the terms of the license, it is extended to all recipients, even if SE says otherwise! Regardless of whether they explicitly grant the CC license to 3rd parties, or specifically offer the data under another license: any recipient of the CC licensed content, even if only in the form of "adapted materials", still also gets an unrestricted and royalty-free CC-BY-SA license, just by virtue of receipt. If SE doesn't inform the recipient of the original CC-BY-SA license, they are in violation of the attribution section of the terms. If they do, then the 3rd party recipient doesn't need SE's permission to claim or use the CC license themselves. They simply have it, automatically.
As an aside, the license also prohibits SE from charging any kind of "royalty" for access to the content itself, which would arguably include most efforts to fully "guardrail" commercial data access. They can charge for other things besides access to the content, for instance preparing the data in a special format that's more portable, or allowing someone to run larger queries against the live database. But this is true only if they share the same data freely in some other manner. Otherwise any "monetization" plans can be construed as data access charges, thus violating the license.
Moreover, as long as they are sharing the licensed content at all, they are actually supposed to keep a URL link to it, along with notation of edits. There is some flexibility in the wording, and it's not a strictly universal requirement, but the exception is intended for data domains where the content is fully available in other forms, and where creating such links would impose an excessive burden. It would be a hard sell for SE to claim providing a link is infeasible, given the nature of the site.
How does this impose any requirement to do the data dumps?
Well, just because they (must) provide the content links doesn't mean they want everyone to use them exclusively. That's highly resource intensive, and unnecessary for compliance. They don't have to serve up a whole site page to satisfy the license, they just have to offer a URL that responds with the appropriate content and metadata. As an example, certain API functionality can substitute entirely².
However, they can't add any access control mechanisms which effectively prevent someone from retrieving a "live" piece of licensed content from the designated URL for that content, even if it's the 4723rd one they've asked for today. Someone may need to see that particular post for reference reasons. If the regular HTTP site is the only means of accessing the content, then it must be served to everyone. There is no way for SE to distinguish between allowable and disallowable reasons (if any even exist) for requests. A page load is a page load.
Again, they can take down content entirely at any time. But if they reshare in any form, in any context, they'd have to put the public link back up too. So it's not like they can keep a cache of "hidden gems", deleted in public, and stashed away for the paid version only.
The data dumps are simply the cheapest, most efficient, and most practical way for them to comply with the license terms. If SE continues to do full data dumps, then they have some leeway to rate-limit or otherwise "gate" the HTTP site, and to some extent even the API. If not, then they have to leave the API open, or let us scrape the HTTP site ad nauseam, regardless of performance impacts.
It's really more of a "yes, if..."
The dumps are only an obligation to the extent that other substitute means of accessing the data are hobbled by disallowed restrictions. There is no intrinsic requirement to maintain them, but if SE executes its intended plans to limit commercial access elsewhere, they will become required, as the only remaining license-compliant source for the content. In fact, depending on the nature of those other guardrails, they might even be required to increase the frequency of dumps, just to be safe.
All this is to say, they can opt out of the data dumps specifically, but probably not while simultaneously limiting HTTP requests - at least not without providing an alternative. The data dumps are their "get out of jail free" card. If someone complains about rate limits on the HTTP site, or certain restrictions on the API, they can be referred to the dump. And if someone needs to validate the dump, then minimal API/SEDE lookups - or at worst some light scraping - should usually suffice. More broadly, publishing the dump creates conditions in which they can fairly assert certain desired "guardrails" on other data channels without running afoul of the license terms.
Did they actually violate the license this time?
In the present case, the question of whether SE violated the license is moot, because they rectified the issue within the 30-day grace period allowed by the license terms. However, we can still examine it academically. All we need ask is: was the data ever unavailable without disallowed restrictions? Obviously, the HTTP site currently has impermissible "technological measures" which limit access, and thus cannot satisfy the license. So then it hinges upon whether API access was restricted at any point in early June (I don't know if it was). If they didn't lock down the API at all during that time, then the licensed content always remained fully available (at a URL, no less). In that case, they didn't violate the license. Otherwise - they probably did.
¹There is an argument to be made that past promises to always share dumps created an implicit contract, and that content created/shared before a public announcement ending them was only licensed under CC-BY-SA contingent upon fulfillment of that contract. That potentially justifies an argument for voiding the ToS contract which granted the license, and thus a "backdoor" means of rescinding the (otherwise perpetual) CC-BY-SA license even if they don't violate it directly. However, they could avoid that issue simply by announcing (with sufficient notice) that their last dump will be the last dump.
²Note that they don't have to offer the API either. It's just that offering open API access likely satisfies the license, thus freeing them to restrict or limit other forms of access.
