fix(builtins): awk array assignment fails with field refs and compound subscripts

[chaliy]

Bug

Multiple awk array features are broken, surfaced by eval tasks config_ini_merge, db_csv_join_aggregate, build_script_generator, and config_env_template. All 5 eval models hit these bugs.

1. Array assignment with field references as indices

category[$1] = $3 fails with awk: invalid assignment target. The parser at awk.rs:1020-1082 doesn't recognize FuncCall("__array_access", ...) when field references ($N) are used as subscripts.

echo "a b c" | awk '{ arr[$1] = $3; print arr["a"] }'
# Expected: c
# Actual: awk: invalid assignment target

2. Multi-subscript arr[e1, e2] syntax unsupported

POSIX awk a["x","y"] (comma-separated subscripts, concatenated with SUBSEP) fails with expected ']'. The parser at awk.rs:1562 calls parse_expression() which doesn't handle commas as SUBSEP separators inside [].

echo "test" | awk 'BEGIN { a["x","y"] = 1; print a["x","y"] }'
# Expected: 1
# Actual: awk: expected ']'

3. SUBSEP variable undefined

The POSIX awk SUBSEP variable (default \034) is never initialized. Grep for "SUBSEP" in awk.rs returns zero matches.

4. Pre-increment of array element ++arr[key] unsupported

++arr[key] fails because the PreIncrement handler at awk.rs:1353 only accepts AwkExpr::Variable(name), but arr[key] parses as FuncCall("__array_access", ...).

echo "a\nb\na" | awk '{ ++count[$1] } END { for (k in count) print k, count[k] }'
# Expected: a 2\nb 1
# Actual: error

Impact

These bugs make any multi-file awk processing with associative arrays impossible. Eval tasks requiring CSV joins, INI merging, or dependency graph processing all fail across all 5 models.

Eval Tasks Affected

• config_ini_merge (all 5 models FAIL)
• db_csv_join_aggregate (Haiku, Opus FAIL)
• build_script_generator (4/5 FAIL)
• config_env_template (GPT-5.2, GPT-5.3-Codex FAIL)