Handle directory-open errors through on_error during tree walks

[shkyyy18]

Summary

iter_tree_entries() and iter_tree_files() document on_error as the filesystem-error handler (and None as ignoring filesystem errors), but an error opening a directory with os.scandir() bypasses the handler. A directory removed before scanning or an unreadable child therefore aborts the entire walk, even when the caller supplies a handler to log and continue.

from pathspec.util import iter_tree_files

errors = []
# Use a path that does not exist:
list(iter_tree_files("missing-directory", on_error=errors.append))
# Before: FileNotFoundError; errors stays empty.
# After: []; errors contains the FileNotFoundError.

Catch errors from opening each directory in both walkers, remove that directory from the active-ancestor memo, call the handler when present, and return from that subtree. Exceptions raised by the handler still propagate. Successful scanning and recursive-loop detection are unchanged. This is intentionally limited to opening a directory, not a general rewrite of entry inspection or iteration.

Tests

• Added missing-root tests for ignore, record, and re-raise behavior in both walkers.
• Added synthetic permission-denied child tests for both walkers; healthy siblings remain discoverable. These inject PermissionError at os.scandir, rather than relying on platform-specific permission manipulation.
• Before the fix: all 10 new subcases error. After: both new test methods and all 10 subcases pass.
• Python 3.12 / Windows: python -m unittest discover -t . -s tests/ is OK (225 test methods; 398 skipped cases/subcases). Unmodified baseline: 223 methods, same 398 skips. Optional re2/hyperscan backends and unsupported symlink cases are skipped; no full backend/platform matrix or docs-build claim.
• git diff --check passes. Updated changelog source and generated changelog together.

AI assistance disclosure: this patch and its regression tests were developed with OpenAI Codex assistance; the behavior above was reproduced and tested locally.