Ignore expanded entity nodes in source-backed sniffs

[NickSdot]

Prevents sniffs from reporting violations for DOM nodes introduced by entity expansion rather than present in the inspected source file.

Depends on #29, #30, and #31.
This PR diff will shrink as preceding PRs are merged and this branch is rebased.

[NickSdot]

Rebased and ready for review.

[jordikroon]

Before:
FOUND 114632 violation(s) (114632 error(s), 0 warning(s)) in 10370 file(s).

After:
FOUND 50759 violation(s) (50759 error(s), 0 warning(s)) in 9993 file(s).

Given that #33 will introduce --wide, this behaviour is expected. But good to know.

[jordikroon]

While it's debatable if they are unit tests. Could you merge them into tests/Unit instead?

[NickSdot]

Sure, can follow your preference! Before I move them back: the end result in later branches is having Unit, Feature, and Integration. That is because there will be tests that run actual git commands for e2e testing. Given that, still all to Unit?

Ref: https://github.com/NickSdot/php__docbook-cs/tree/stack-09-performance/tests

[jordikroon]

Ideally, yes. As it would also be more easy to distinguish between tests and non-tests (fixtures & Support).

[NickSdot]

Before: FOUND 114632 violation(s) (114632 error(s), 0 warning(s)) in 10370 file(s).

After: FOUND 50759 violation(s) (50759 error(s), 0 warning(s)) in 9993 file(s).

Given that #33 will introduce --wide, this behaviour is expected. But good to know.

Ah, I should have mentioned this here. This is a change in behaviour that is not only related to --wide. Assume shared has violation on <foo> which is used in A and B, current main reports the same violation for A and B, although it only can be fixed in shared. The current behaviour in main is similar to what PHPStan does with traits; but in our situation that is not actually helpful coz a consumer cannot fix the violation by changing it's own code (unlike traits reported by stan). After this change these kind of violations are reported only once against shared in full or --wide runs. This gets relevant when fixers come in, coz a fixer cannot edit <foo> in expanded A coz <foo> exists in shared; having less overall reports is a side effect.