Python Multi-Log Merge and Timeline Correlator Prompt

You are an incident responder who reconstructs what happened by stitching logs from several services into one timeline, and you build a tool that does it correctly across mismatched timestamp formats and time zones.

I will provide:
- The log files to merge and each file's timestamp format (or a note that it must be auto-detected)
- The time zone of each source and the target time zone for the merged timeline
- Optional filters: a time window, severity threshold, or a correlation key (request ID, trace ID) to follow across services

Your job:

1. **Parse timestamps robustly** — support per-file strptime formats and common ISO-8601 variants, normalize every line to timezone-aware UTC, and handle lines with no parseable timestamp (attach them to the previous timestamped line as a continuation, e.g. multi-line tracebacks).
2. **Merge efficiently** — stream files line by line and merge with a heap/`heapq.merge` on timestamp rather than loading everything into memory, so multi-gigabyte logs work.
3. **Tag provenance** — prefix or annotate every emitted line with its source file/service so the merged timeline is unambiguous.
4. **Correlate** — when given a correlation key (request/trace ID), filter the timeline to just the events carrying that key across all sources, preserving order.
5. **Window and filter** — support a `--since/--until` window and a minimum severity, applied during the streaming merge.
6. **Output** — emit a clean, aligned, time-ordered timeline for humans and optional JSONL for further processing.
7. **Test** — pytest cases for mixed timestamp formats, time-zone normalization, multi-line continuation, and correlation-key filtering.

Output as: (a) the per-source parser/normalizer, (b) the heap-based streaming merger, (c) the CLI with window/severity/correlation filters, (d) the pytest suite.

Bias toward: timezone-aware UTC normalization, streaming/heap merge over in-memory sort, correct multi-line record handling, and unambiguous source tagging.