Let a trait-pathed ignore entry take precedence over the class-pathed ones

[mlocati]

When a class using a trait reports its own error with the same message and identifier as an error deduplicated directly into the trait, --generate-baseline records the class occurrence under the class path (count: 1) and the deduplicated error under the trait path (count: 1)
On re-run the class-pathed entry also absorbed the trait context of that class, so it was reported as expected 1 time but occurred 2 times: the generated baseline failed by itself.

An entry scoped to the trait file hides the deduplicated error for every using class, so it is now tried first: the class-pathed entries account only for the errors reported in the class files themselves.

[mlocati]

The problem this PR fixes, in short: a freshly generated baseline fails on the very next run.

With the provided test files, AMailer and BMailer use SomeTrait, and both AMailer and SomeTrait contain the same error. phpstan analyse reports 2 errors:

src/AMailer.php:19: Right side of && is always true.
src/SomeTrait.php:17: Right side of && is always true.

(the trait error is reported once, although the trait is used by two classes).

phpstan analyse --generate-baseline=baseline.neon accordingly writes 2 entries, both with count: 1: one for src/AMailer.php, one for src/SomeTrait.php.

Then phpstan analyse -c with-baseline.neon fails without this patch:

src/AMailer.php:19: Ignored error pattern ... in path src/AMailer.php is expected to occur 1 time, but occurred 2 times.

because the trait error, when checked against the ignore entries, is also matched as an error of AMailer (the class using the trait), so the AMailer.php entry gets charged 2 errors instead of 1.

With this patch, the SomeTrait.php entry is checked first, so the trait error is matched by it, the AMailer.php entry only matches its own error, and the run passes.