Remove an optimization on PathSelector producing false negatives

[desruisseaux]

Remove an optimization on PathSelector which was sometime producing false negatives. It is safe to use an exclude filter for directories during FileVisitor transversal, but it is unsafe to use an include filter because of the need to traverse parents first.

[gnodet]

Review: APPROVE ✅

Well-reasoned fix that removes the unsafe include-based directory pre-filtering optimization from DirectoryPrefiltering. The rationale is sound: excluding directories is safe (if a directory matches an exclude, all files within are excluded), but include-based filtering is inherently unsafe because FileVisitor must enter parent directories before discovering matching children. For example, with include pattern **/*foo*/**, the old code would reject directory org/ because it doesn't contain "foo", preventing discovery of org/foo/ and files beneath it.

The fix correctly:

• Removes the ignoreIncludes field and all include-based logic
• Simplifies matches() to accept any directory not explicitly excluded
• Updates simplify() to return INCLUDES_ALL when no excludes exist
• Flips the appropriate test assertions with clear explanatory comments

The trade-off (slightly more directory traversal for correctness) is the right call, especially since file-level includes/excludes remain unchanged. The exclude-based filtering (retained) handles the most common performance case (skipping .git, target, etc.).

Minor nit: A couple of small language issues in the test comments — "Before to match" → "In order to match" / "Before matching", and "Same reason than above" → "Same reason as above". These are cosmetic only.

---

Reviewed via automated review loop · Feedback? Tag @gnodet