[pganssle]

Part of #107862. This adds Lib/test/test_codecs_property.py, Hypothesis property tests for the codec machinery, in the same shape as #159960 for the compression modules.

The existing codec tests check each codec against fixed strings, and test_basics runs the incremental and stream interfaces on "abc123" only. The new file runs every codec in all_unicode_encodings on generated input and compares against the stateless codec functions:

• incremental encode and decode for any chunking of the input, under every error handler, and getstate()/setstate() round trips (the conditions documented for IncrementalDecoder.getstate());
• StreamWriter then StreamReader with arbitrary write and read sizes, and readline()/read() interleaving against str.splitlines();
• the error handlers run back through the codec: backslashreplace output decoded via unicode_escape, xmlcharrefreplace via html.unescape, surrogateescape round trips on the ASCII-compatible single-byte codecs;
• charmap_decode/charmap_encode with generated tables against a small oracle, and charmap_build maps against equivalent dicts;
• encode/decode round trips, escape_encode/escape_decode, the escape codecs with surrogates, the bytes transform codecs and rot_13, and name normalization in codecs.lookup.

With Hypothesis the file takes about five seconds; 3000 examples per test also pass. Without it the explicit examples replay through the stubs (27 tests, 6 expected failures in both modes).

The by-design differences are written down in the module rather than hidden in the strategies: BOM-less utf-16/32 is rejected incrementally (tested as such), utf-7 accepts lone surrogates and closes its shift sequence at every chunk, cp932/cp950/euc_jp/shift_jis fold a few compatibility characters, idna applies nameprep, raw_unicode_escape leaves backslashes alone, and surrogateescape only round-trips on codecs where every byte decodes injectively.

Six real divergences turned up, each pinned as an expectedFailure with the minimal input as an explicit @example and carved out of the broad tests only where the comments say so. Two are already on the tracker; issues for the others to follow:

1. The incremental decoders of euc_jp, euc_jis_2004, euc_jisx0213 and gb18030 drop input after an incomplete trailing sequence under surrogateescape: b'\x806'.decode('gb18030', 'surrogateescape') is '\udc806', the incremental decoder gives '\udc80' (also through TextIOWrapper).
2. The ISO-2022 encoders write U+001B as a bare ESC byte that their decoders reject or misread ('\x1b\xa0'.encode('iso2022_jp_2004').decode('iso2022_jp_2004') is '\x1b\x1b$(Q)"'); iso2022_kr does the same with SO and SI.
3. codecs.StreamReader.read() never decodes with final=True, so an incomplete trailing character is silently dropped: codecs.getreader('utf-8')(io.BytesIO(b'ab\xe2')).read() is 'ab', with no error in strict mode and no replacement character in replace mode. This is codecs.StreamReader doesn't pass final=1 to the UTF-8 codec #56717, open since 2011.
4. The punycode incremental codecs transform each chunk on its own ('a' then 'b' encodes to b'a-b-', not b'ab-'). test_codecs already lists punycode as "broken in stateful mode". Many incremental codecs don’t handle fragmented data #64331 mentions it in passing.
5. '\u3164'.encode('euc_kr') gives the Hangul filler b'\xa4\xd4', which euc_kr cannot decode on its own (cp949 can). This was euc_kr char '0x3164' decode('ksx1001') cause UnicodeDecodeError  #101863, closed as not a bug; the test records the asymmetry either way.
6. iso2022_jp_3 encodes U+9B1D but decodes the result as U+9B1C: in jisx0213_2000_2_decoder the 2000-table emulation macro is followed by if instead of else if, so the table lookup overwrites it.

• Issue: [Meta] Research: what can we test with Hypothesis? #107862