Fix panic in subset when glyph_ids maps to only .notdef

[MasKusuno]

Summary

• When subset() is called with CmapTarget::Unicode and glyph IDs that have no cmap mapping (e.g., only glyph ID 0 / .notdef), MappingsToKeep ends up empty. This causes CmapSubtableFormat4::from_mappings to panic on an unconditional unwrap() of the first mapping iterator element.
• In WASM builds this manifests as an "unreachable" trap, making it difficult to diagnose.
• This is easily triggered in practice when subsetting a font with codepoints it does not contain — common in CJK font fallback chains where different fonts cover different Unicode ranges.

Changes

• Return SubsetError::NoGlyphs early in subset() when mappings are empty and the cmap target is Unicode
• Replace the unwrap() in CmapSubtableFormat4::from_mappings with Err(ParseError::BadValue) as defense-in-depth
• Add is_empty() method to MappingsToKeep<T>
• Add NoGlyphs variant to SubsetError
• Add two tests (CFF/OTF and TTF) verifying .notdef-only input returns NoGlyphs instead of panicking

Test plan

• cargo test --lib — 336 passed, 0 failed
• cargo test (all integration + doc tests) — all passed
• cargo fmt -- --check — clean

Context

Discovered while building a WASM-based CJK font subsetter for GJS Kanji Database. When a codemap requests a font chunk for a Unicode range that the font doesn't cover, the subsetter receives only .notdef glyph IDs and crashes.

[wezm]

Good find, but I'm not sure the fix is the right approach though. In a scenario where only glyphs that do not have a cmap entry are retained, MappingsToKeep legitimately ends up empty. With the proposed changes an error would be returned. Instead I think that it should still build a cmap table. I.e. a cmap that is present, but has fields populated such that no glyphs are mapped.

[MasKusuno]

Thanks for the review! You're right — returning an error is the wrong approach since empty mappings is a legitimate case.

I've updated the PR to build a valid cmap table with no glyph mappings instead:

• CmapSubtableFormat4::from_mappings: when mappings are empty, produces a table with only the required 0xFFFF sentinel segment
• CmapSubtableFormat12::from_mappings: when mappings are empty, produces a table with an empty groups list
• Removed the SubsetError::NoGlyphs variant and the early-return guard in subset()
• Updated the tests to verify that .notdef-only input produces a valid, parseable font

All 336 lib tests + 13 doc tests pass, cargo fmt clean.

[wezm]

Looks good overall. Just a little more iteration needed on both of the tests.

Also please squash all commits into a single commit when you're done.

[wezm]

This will not parse much of the font. Instead the cmap table should be parsed like test_subset_with_macroman_cmap and test_subset_with_os2_and_unicode_cmap.

[wezm]

We're already in a subset module, so the subset_ prefix can be removed from the test names.

[MasKusuno]

Thanks for the feedback! I've updated the tests to properly parse the cmap table (matching the pattern in test_subset_with_macroman_cmap / test_subset_with_os2_and_unicode_cmap), removed the subset_ prefix from test names, and squashed everything into a single commit.

[wezm]

Applied in 5fb3fff