feat(bindings): mirror TextEncoding

[andiwand]

🤖 Generated with Claude Code

PR 7 of the v7 API plan.
Stacked on #836. Non-breaking — this only adds.

Why this exists

It was not in the first draft of the plan. #833 turned it up: TextEncoding is
mirrored in no binding. Java, python and objc each build a charset()
string from TextFile::encoding() and stop there, so from outside C++ you
could read the name of an encoding you had already been handed, and nothing
else — not enumerate them, not resolve an alias, not ask whether one is
actually decodable as opposed to merely named.

That is a gap on its own, and it is also a blocker: CsvOptions carries a
TextEncoding, so PR 8 cannot fold it into DecodeOptions until the enum
crosses.

What lands

The enum plus all five lookups, in every binding:

Java	TextEncoding enum, canonicalName(), names(), isDecodable(), byName(), all(); TextFile.encoding()
Python	pyodr.TextEncoding, all_text_encodings, text_encoding_to_string, text_encoding_by_name, text_encoding_names, text_encoding_is_decodable; TextFile.encoding()
ObjC	ODRTextEncoding, Odr.allTextEncodings / string(textEncoding:) / textEncoding(name:) / names(textEncoding:) / isDecodable(textEncoding:); TextFile.encoding
wasm	Odr.enums.TextEncoding

The 41 enumerators were generated from file.hpp rather than retyped, so the
ordinals cannot drift by transcription.

charset() stays everywhere it was, now layered on encoding(). wasm derives
its table from all_text_encodings() the way FileType already does, instead
of restating ordinals — the one hand-written part is unknown, which
all_text_encodings() deliberately leaves out because it has no name.

Tests

New in all four: name round-trips including alias resolution (CP1252 →
windows-1252), unknown having no name, decodable being narrower than
named (shift_jis is named, not decoded), all() excluding unknown, and a
text file reporting its encoding. The wasm case is pinned in enums.test.mjs
beside the existing ordinal guards.

Verified

Full build; the JNI junit suite, 71 python tests and the C++ suite pass.
Apple (11 TUs) and wasm (6 TUs) syntax-checked with -Wall -Wextra -Werror
against their real toolchains and deployment targets — see the note on #836
about why that check is now part of the loop here.