fix(cli): reject non-finite --timeout values

[twelfthlabor]

Part of #3120.

Problem

timeout_check only rejects values <= 0. --timeout inf and --timeout nan pass validation and crash the entire run with unhandled exceptions from the socket layer (OverflowError: timestamp out of range for platform time_t / ValueError: Invalid value NaN), escaping get_response's except clauses. Huge finite values like --timeout 1e5 are accepted (user's explicit choice, intentionally not capped here).

Fix

if not math.isfinite(float_value) or float_value <= 0:
    raise ArgumentTypeError(...)

nan previously slipped even the <= 0 check (nan <= 0 is False). Large finite values remain accepted; only non-finite values are rejected. The "huge finite values block indefinitely" half of #3120's title stays out of scope: capping large finite values would override the user's explicit choice, so the issue stays open for it.

Relation to open PRs #2876 / #2867 / #2955

All three address #2866 (non-numeric input) and keep if float_value <= 0 unchanged, so inf/nan pass through all of them. This is complementary; no conflict.

Testing

• New test test_timeout_check_rejects_nonfinite_values: fails on master (DID NOT RAISE), passes here; '1e5' acceptance pinned as a regression guard.
• End-to-end (offline, stubbed network): --timeout 1e5 → 100000.0 reaches the query layer; inf/nan/-5 → clean SystemExit(2) instead of a crash.
• Full offline suite: pytest tests -m "not online" -q → 16 passed
• ruff: findings identical to master (single-line insertion shift only)

Verification

Independently re-verified on fresh checkout: condition reviewed line-by-line, test proven to bite on master, success paths unchanged end-to-end, ' 1.5 ' whitespace behavior unchanged.