extract_metric() and extract_trace_payload() both called
json.loads(event["message"]) independently. For plain log events (the
majority), both fail with exceptions — 2x parse + 2x exception
overhead per event.

Parse once in split() via _try_parse_message() and pass the result to
both extractors through a parsed_message parameter. A _PARSE_FAILED
sentinel avoids conflating parse failure with a missing argument.

Short-circuit in split(): when a metric is found, skip the trace check
entirely. This is both more efficient and avoids a shared-dict mutation
issue where extract_metric appends to the parsed dict's "t" list.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Address review feedback: instead of threading a parsed_message param
through extract_metric and extract_trace_payload, parse once in split()
and pass the parsed dict directly.

- split() handles json.loads and classifies parse failures as logs
- extract_metric(parsed, event) validates and enriches a pre-parsed dict
- is_trace(parsed) is a pure predicate — the trace payload dict is
  built in split() since it only needs the raw message and tags
- Removes _PARSE_FAILED sentinel and _try_parse_message helper

Tests updated to match the new signatures with better coverage of
actual validation logic.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>