[5.x] Fix param leak in AdminUrlGenerator::generateUrl

[ousamabenyounes]

Q	A
Branch?	5.x
Bug fix?	yes
New feature?	no
Deprecations?	no
Issues	Fix #7734
License	MIT

AdminUrlGenerator::generateUrl() ends with a state reset whose own comment
states it is "important to start the generation of each URL from the same
initial state, otherwise some parameters used when generating some URL could
leak to other URLs". But three earlier return statements — the empty-route
dashboard link and both EA::ROUTE_NAME branches (the linkToRoute() path) —
returned before that reset ever ran.

As a result, once a URL was generated through the linkToRoute() path, the
generator stayed initialized with routeName/routeParams still set. The next
URL generated in the same request (e.g. a linkToCrudAction() global action)
re-entered the ROUTE_NAME branch and lost its filters/page/sort/query
parameters — even though every other link on the page kept them.

The fix moves the reset into a finally block so it runs on every return path
(and on exceptions), which is exactly the "same initial state" guarantee the
existing comment promised.

Test verification (RED → GREEN)

New test AdminUrlGeneratorTest::testRouteParametersDontLeakToNextUrl generates
a custom-route URL and then asserts the following URL is clean.

RED — on 5.x with only the test applied (before the fix):

1) EasyCorp\Bundle\EasyAdminBundle\Tests\Unit\Router\AdminUrlGeneratorTest::testRouteParametersDontLeakToNextUrl
Failed asserting that two strings are identical.
--- Expected
+++ Actual
@@ @@
-'http://localhost/admin?foo=bar'
+'http://localhost/admin?routeName=some_route&routeParams%5Bkey%5D=value'

FAILURES!
Tests: 1, Assertions: 1, Failures: 1.

GREEN — with the fix applied, whole test class:

Testing EasyCorp\Bundle\EasyAdminBundle\Tests\Unit\Router\AdminUrlGeneratorTest
..................                                                18 / 18 (100%)

OK (18 tests, 35 assertions)

phpstan analyse src/Router/AdminUrlGenerator.php and php-cs-fixer (dry-run)
both report no issues on the changed files.

[ousamabenyounes]

Note on CI: the single red job phpunit (ubuntu-latest, 8.4, 8.0.*) is
unrelated to this change. It fails suite-wide (~400 errors) on functional
tests such as ImageFieldTest, which this PR doesn't touch. Running the
unmodified 5.x tip against symfony/*:8.0.* reproduces the same failures
(e.g. ImageFieldTest 8/8 failing with "actual size 0 matches expected
size 1"), so it looks like a Symfony 8.0 point-release compatibility break
in the test app rather than a regression here. This PR only touches
AdminUrlGenerator, whose unit test passes on every green matrix row.