Report the comma position for illegal trailing comma errors - #382

Differential fuzzing of the C scanner against the pure-Python fallback (_toggle_speedups) found 61 divergences in 12,000 random decode inputs, all of them the position reported for an illegal trailing comma in an array:

>>> simplejson.loads('[42, ]')
# C scanner:      ... line 1 column 6 (char 5)   <- the ']'
# Python scanner: ... line 1 column 5 (char 4)   <- the space before it
# stdlib json:    ... line 1 column 4 (char 3)   <- the comma

Cause

ERR_TRAILING_COMMA_* is raised after the whitespace following the comma has been skipped, so the index passed no longer refers to it. The C scanner passes idx, the closing bracket. JSONArray passes end - 1, an idiom copied from the delimiter error a few lines up where end had been incremented; here it has not, so the reported character is whatever precedes ].

Fix

Capture the comma index when the delimiter is confirmed and report that, as json/decoder.py does with its own comma_idx. Four sites: JSONObject, JSONArray, and the object and array branches of the scan template. Trailing-comma detection landed in 4.0.0 as CPython parity (#371), so stdlib breaks the tie.

Result

Same corpus: 61 divergences before, 0 after. Refiltering it to drop trailing commas gives 0 both before and after. Encode was fuzzed alongside and showed 0.

test_trailing_comma_position pins the offsets for both containers, whitespace after the comma, and nesting, and _cibw_runner runs it on both paths. Reverting only decoder.py fails it on the Python path, only _speedups_scan.h on the C path. Suite passes on both (460 tests, was 458). The C change builds clean under the -Werror flags in AGENTS.md. I could not exercise the Py2 _str instantiation locally.