[upstream] Move FormatSetter from middleware to a route filter, removing conditionality on Accept header

Oh, that sounds lovely! 👏 I think we'd all love to remove this middleware!

I fixed the test failures over at #3029373-6: PageCache::getCacheId() sometimes uses different cache IDs during get and set, causing unnecessary cache misses for some requests.

With both that core patch applied and the JSON:API patch that I attached here, tests are mostly passing:

Testing Drupal\Tests\jsonapi\Functional\NodeTest
.......F...                                                       11 / 11 (100%)

Time: 1.95 minutes, Memory: 6.00MB

There was 1 failure:

1) Drupal\Tests\jsonapi\Functional\NodeTest::testPostIndividual
Failed asserting that Array &0 (
    0 => 'text/html; charset=UTF-8'
) is identical to Array &0 (
    0 => 'application/vnd.api+json'
).

So it looks like this will be viable :)

I dug in and that failing test is legitimate. Some error responses happen before or during routing, and at that time the JSON:API format hasn't been negotiated yet. So … you get a HTML response.

This may be acceptable.

#3029373: PageCache::getCacheId() sometimes uses different cache IDs during get and set, causing unnecessary cache misses for some requests just landed! Retesting.

Oh, right, this can never be committed to JSON:API 2.x since it must work for 8.5+8.6+8.7. If we're lucky, we could get #3029373: PageCache::getCacheId() sometimes uses different cache IDs during get and set, causing unnecessary cache misses for some requests backported to 8.6. But it'll never get backported to 8.5.

So, we can only do this as part of #3015325: [ignore] support issue for the core patch AFAICT.

For the reasons given in #14, this passed tests on 8.7, but not on 8.5 or 8.6.

Thoughts?

#19: done: #3015325-31: [ignore] support issue for the core patch.