Invert the font cmap once when subsetting

[danmolitor]

Addresses #182. Thanks @captyoungkyu for the precise diagnosis and patch; this is the same approach.

Font subsetting found each glyph's codepoint by scanning 0..=0xFFFF per glyph. Hangul starts at U+AC00, and a glyph with no codepoint (a ligature or shaped form) scanned the whole plane, so serialization time grew with the number of distinct Korean syllables. The glyph-to-codepoint map is now built in one pass over the font's Unicode cmap, keeping each glyph's lowest BMP codepoint as the scan did, so subset fonts are unchanged.

Measured with Noto Sans KR, one page of Hangul (native CLI, before / after):

Distinct syllables	before	after	output
100	720 ms	262 ms	byte-identical
1,000	4,662 ms	85 ms	byte-identical

• New test: the one-pass map against the full-plane scan, for every glyph in six fonts (Noto Sans, Noto Emoji, Noto Naskh Arabic, Noto Sans Devanagari, Noto Sans Hebrew, Liberation Sans). Checked that it catches a wrong mapping (keeping the highest codepoint instead fails it). .notdef is excluded: a format 4 table's final segment maps U+FFFF to it, and the subsetter never gives it a cmap entry.
• A timing test would be flaky in CI, so the before/after above is measured rather than asserted.
• Engine (671), visual, html and package tests pass; the byte wall is identical; veraPDF passes PDF/UA-1 and every PDF/A level. Gallery manifest regenerated (hashes only).

[captyoungkyu]

Thanks for the quick turnaround and the thorough verification!

I'll test it on our invoices once it's released.