fix: avoid unicode filepath suffix panic

[tmdgusya]

Summary

• fix a remaining UTF-8 boundary panic in path_ends_with_suffix()
• return false instead of panicking when a byte-derived suffix offset lands inside a multibyte character
• add regression tests for the helper and for apply_constraints(Constraint::FilePath(...))

Similar PR / duplicate check

I checked existing PRs before opening this:

• closest prior fix: fix: Unicode segmentation crash #373 (fix: Unicode segmentation crash)
• no open PR currently covers this remaining Constraint::FilePath / path_ends_with_suffix() panic path

This PR is intentionally narrow: it fixes the unchecked path[start..] slice in crates/fff-core/src/constraints.rs without changing matching semantics.

Root cause

path_ends_with_suffix() computed:

let start = path.len() - suffix.len();

and then sliced with:

path[start..]

start is a byte offset, not guaranteed to be a UTF-8 char boundary. For Unicode filenames, a non-matching suffix can make start land inside a multibyte codepoint, which panics before constraint filtering can return false.

Fix

Use path.get(start..) instead of unchecked indexing:

• if start is not a valid char boundary, return false
• otherwise preserve the existing ASCII-insensitive suffix comparison and / boundary behavior

Verification

I avoided using any user-specific filename in tests and instead used synthetic Unicode fixture names.

Reproduction guard added

New tests in crates/fff-core/src/constraints.rs:

• test_path_ends_with_suffix_does_not_panic_on_unicode_suffix
• test_apply_constraints_file_path_with_unicode_suffix
• test_path_contains_segment_does_not_panic_on_unicode_segment

The important regression case uses a synthetic filename such as:

• data/유니코드_파일_테스트.csv

and a deliberately non-matching suffix that would previously place the byte offset in the middle of a multibyte character.

Commands run

cargo test -p fff-search constraints::tests -- --nocapture

Result

All constraint tests pass locally after the fix, including the new Unicode regression coverage.

Scope notes

• no parser changes
• no matching behavior expansion
• no Unicode normalization/case-folding changes
• only panic prevention for valid UTF-8 input in the file path suffix constraint path

[dmtrKovalenko]

@copilot resolve the merge conflicts in this pull request