test: update literal types after template union fix

[calebdw]

No description provided.

[phpstan-bot]

You've opened the pull request against the latest branch 2.3.x. PHPStan 2.3 is not going to be released for months. If your code is relevant on 2.2.x and you want it to be released sooner, please rebase your pull request and change its target to 2.2.x.

[staabm]

@ondrejmirtes its a 2.3.x only change in expectations. Please have a look

[SanderMuller]

I am not the maintainer, so the semantics call here is Ondřej's. What I can add is where the change comes
from, because it is not one of the template commits on 2.3.x.

The transcription is faithful. On upstream/2.3.x the analyser produces exactly the three values this
PR writes down:

Expected: Bug13374\Collection<int|string, int|string>   Actual: Bug13374\Collection<int|string, 'foo'|int>
Expected: Bug13374\Collection<int|string, int>          Actual: Bug13374\Collection<123|string, int>
Expected: Bug13374\Collection<int|string, int|string>   Actual: Bug13374\Collection<123|789|string, 'foo'|int>

The test fails on 2.3.x without the PR and passes with it. On 2.2.x it passes as it stands, so "2.3.x only"
is right.

Where it started. I bisected 2.2.x..2.3.x with the test file pinned, so the oracle stayed the same at
every step. The first commit with the new behaviour is d9bba3628, the last merge of 2.2.x into 2.3.x, and
that merge brought exactly one commit: ea3e1126d, #6376. That is the PR that created this test file. So
this is #6376's new test meeting the template work on 2.3.x, not a regression from a single 2.3.x commit.

The new values follow from the annotations. With @phpstan-this-out static<TKey|TPutKey, TValue|TPutValue>
and a receiver that keeps its mutated type:

call	this-out	result
push(123)	int|string, int|123	int|123 collapses to int
push('foo')	int|string, int|'foo'	'foo'|int
put(123, 456)	string|123, int|456	123|string, int
put(789, 'foo')	123|string|789, int|'foo'	123|789|string, 'foo'|int

That also explains why only three of the four assertions move. In the first one 123 is absorbed by int,
so it reads the same under either behaviour and cannot tell them apart.

One thing I checked and it is not the cause. This is not a general shift toward literal template
arguments. A plain @template T @param T $v @return T returns 'foo', and @template TK of array-key
returns 123, on both branches. The difference is narrower: 2.2.x generalises the TValue|TPutValue
union that @phpstan-this-out composes, and 2.3.x keeps it literal.

Worth a decision rather than a review comment. The key accumulates across sequential calls, so
123|789|string after two put() calls. In straight-line code each further call adds another literal. If
this-out should generalise, then the fix belongs in the inference and this PR would be pinning the wrong
values.

Nothing to measure here, the PR touches no src/. All three red checks belong to the base:
PHPStan (8.1, windows-latest) and the two macos-15-intel Turbo legs are red on #6392, #6398 and #6403
as well.

[ondrejmirtes]

Thank you!