Expand to twelve single-byte encodings via codegen; match msdos-encodings API

Rework the crate on the msdos-encodings design: vendored Unicode
Consortium VENDORS/APPLE mapping files in data/apple/, a
tools/gen_tables.py generator (PEP 723 / uv run / rustfmt) emitting full
[Option<char>; 256] tables, and a fallible decode / decode_lossy /
encode API with non_exhaustive thiserror errors.

Encodings: Roman, Greek, Cyrillic, Central European Roman, Turkish,
Croatian, Icelandic, Romanian, Celtic, Gaelic, Ukrainian (pre-9.0
Cyrillic variant), Inuit. The eight encodings Apple revised for the
Euro carry a Revision payload (Classic = pre-8.5/9.0/9.2.2), generated
from per-encoding deltas with invariant checks that fail regeneration
if the vendored data drifts.

Apple-specific API points beyond the msdos shape: from_text_encoding
(kTextEncodingMac… base values, the HFS+ hint), from_script_code plus
from_fd_script (bit-7 validity flag on the raw Finder Info byte), and
decode_pstr for length-prefixed Str27/Str31 name fields with a
TruncatedPascalString error for corrupt lengths.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

README.md

@@ -4,28 +4,84 @@ Bidirectional, emulator-grade conversion between classic Mac OS text encodings
 and Unicode — without linking ICU. The crate owns the canonical Apple tables, so
 it stays small, self-contained, and cross-compiles cleanly.
 
-```rust
-use apple_encodings::{AppleEncoding, MacRomanRevision};
+These are strictly **Apple's** mappings as published by the Unicode Consortium
+(`VENDORS/APPLE/`), vendored in `data/apple/`.
 
-let enc = AppleEncoding::default(); // Mac OS Roman, post-8.5
-assert_eq!(enc.decode(b"Caf\x8e"), "Café");
+```rust
+use apple_encodings::{AppleEncoding, Revision};
+
+let enc = AppleEncoding::default(); // Mac OS Roman, modern revision
+assert_eq!(enc.decode(b"Caf\x8e").unwrap(), "Café");
 assert_eq!(enc.encode("Café").unwrap(), b"Caf\x8e");
 
-// Pick by the Finder Info `fdScript` byte:
-let enc = AppleEncoding::from_script_code(apple_encodings::SCRIPT_ROMAN).unwrap();
+// Pick by the HFS+ `textEncoding` hint (kTextEncodingMac… base values):
+let enc = AppleEncoding::from_text_encoding(7).unwrap();
+assert_eq!(enc.decode(&[0x80]).unwrap(), "А"); // Mac OS Cyrillic
 
-// Revision matters for exactly one byte (0xDB):
-let classic = AppleEncoding::MacRoman(MacRomanRevision::Classic);
-assert_eq!(classic.decode(&[0xDB]), "¤"); // pre-8.5 currency sign, not €
+// Or by the raw HFS Finder Info `fdScript` byte (bit 7 flags validity;
+// a Cyrillic system writes 0x87). Script system, not table — the Roman
+// regional variants all report script 0:
+let enc = AppleEncoding::from_fd_script(0x87).unwrap();
+assert_eq!(enc.name(), "Mac OS Cyrillic");
+
+// Classic Mac name fields are Pascal strings (Str27/Str31):
+let name = enc.decode_pstr(b"\x08\x84\xEE\xEA\xF3\xEC\xE5\xED\xF2").unwrap();
+assert_eq!(name, "Документ");
+
+// The Euro rollout revised several tables, at different bytes per encoding;
+// pick the pre-Euro table for older-system fidelity:
+let classic = AppleEncoding::MacRoman(Revision::Classic);
+assert_eq!(classic.decode(&[0xDB]).unwrap(), "¤"); // pre-8.5, not €
+
+// Decode is fallible for API uniformity (Greek pre-9.2.2 leaves 0xFF
+// undefined), or go lossy:
+let greek = AppleEncoding::MacGreek(Revision::Classic);
+assert!(greek.decode(&[0xFF]).is_err());
+assert_eq!(greek.decode_lossy(&[0xFF]), "\u{FFFD}");
 ```
 
 ## Status
 
-- **Mac OS Roman** — implemented, both pre- and post-8.5 revisions, decode + encode.
-- **Regional single-byte** (Cyrillic, Greek, Turkish, …) and **CJK double-byte**
-  (Japanese, Big5, GB, Korean) — planned, to be codegen'd from the Unicode
-  Consortium `VENDORS/APPLE/*.TXT` tables. The double-byte tables will be
-  feature-gated.
+- **Twelve single-byte encodings** — implemented, decode + encode: Roman,
+  Greek, Cyrillic, Central European Roman, Turkish, Croatian, Icelandic,
+  Romanian, Celtic, Gaelic, Ukrainian (the pre-9.0 Cyrillic variant), and
+  Inuit.
+- **Pre-Euro revisions** — implemented for the eight encodings Apple revised:
+  Roman/Croatian/Icelandic/Romanian/Celtic/Gaelic (`0xDB`, Mac OS 8.5),
+  Cyrillic (`0xA2`/`0xB6`/`0xFF`, Mac OS 9.0), Greek (`0x9C`/`0xFF`,
+  Mac OS 9.2.2).
+- **CJK double-byte** (Japanese, Chinese, Korean) and the
+  **bidirectional/complex scripts** (Arabic, Hebrew, Thai, Indic — which need
+  multi-scalar decode) — planned, behind the same API; the double-byte tables
+  will be feature-gated.
+
+## Fidelity notes
+
+- Every table is a verified bijection: anything an encoding decodes re-encodes
+  to the identical bytes.
+- Every table is total except Mac OS Greek at `Revision::Classic`, whose
+  `0xFF` was undefined before Mac OS 9.2.2 — so `decode` returns a `Result`
+  and `decode_lossy` substitutes U+FFFD, matching the `msdos-encodings`
+  sister crate.
+- Apple's corporate Private Use Area mappings are preserved, not normalized:
+  the Apple logo (`0xF0` → `U+F8FF` in Roman) and Mac OS Turkish's explicit
+  "undefined character" slot (`0xF5` → `U+F8A0`) round-trip exactly.
+- The pre-Euro revision tables are generated from per-encoding deltas taken
+  from each mapping file's own change history, with invariant checks that
+  fail regeneration loudly if the vendored data ever disagrees — the Euro
+  landed at different bytes per encoding, so nothing is hand-placed.
+- Mac OS Ukrainian is derived from `CYRILLIC.TXT` plus its documented delta
+  (`UKRAINE.TXT` upstream is a notes-only stub: Mac OS 9.0 retired the
+  separate Ukrainian character set).
+
+## Regenerating tables
+
+Everything in `src/tables/` except `mod.rs` is generated from the vendored
+mapping files in `data/apple/` (needs Python ≥ 3.10 and `rustfmt` on `PATH`):
+
+```sh
+uv run tools/gen_tables.py
+```
 
 ## Scope