feat: supervise loader orchestrator failure handling

[Teagan42]

Summary

• replace the loader orchestrator placeholder with a TaskGroup-driven supervisor that cancels related stages, drains queues, and logs failures through mcp_plex.loader.orchestrator
• add regression tests that simulate ingestion and persistence failures to confirm downstream cancellation and cleanup
• bump the project version metadata to 0.26.71 and refresh the lockfile

Why

• structured orchestration is required so pipeline failures do not leave background tasks or queue items hanging, and to emit actionable stage-specific errors

Affects

• loader pipeline orchestration logic, associated tests, and version metadata

Testing

• uv run pytest
• uv run pytest tests/test_loader_orchestrator.py

Documentation

• not updated

---

https://chatgpt.com/codex/tasks/task_e_68e2e8f98b1c832892e14c9f907f3069

[github-actions]

Coverage Report

File	Stmts	Miss	Cover	Missing
mcp_plex/loader
__init__.py	746	129	83%	174–181, 230, 316–321, 923–924, 968, 970, 1035–1036, 1047–1049, 1052, 1055–1056, 1058, 1080, 1090, 1101–1133, 1148, 1150, 1157–1171, 1174–1188, 1194, 1209–1240, 1245–1278, 1281–1285, 1290–1298, 1308, 1376–1378, 1439–1441, 1799
mcp_plex/loader/pipeline
channels.py	73	2	97%	19–20
enrichment.py	334	44	87%	78, 80, 87, 91, 134–137, 170, 191, 211, 219–221, 228–231, 234–236, 244, 301, 357, 380, 384, 386, 412, 430, 465, 471, 474–482, 508, 532, 603, 606–608
ingestion.py	81	8	90%	67, 98–108, 129, 157, 179
orchestrator.py	85	5	94%	48, 112, 141, 164–165
persistence.py	117	9	92%	109, 151–152, 162, 167, 171–173, 201
mcp_plex/server
__init__.py	614	29	95%	43–44, 119–120, 148, 252, 256, 277–280, 297, 362, 365, 402, 420–421, 458, 1109, 1131–1137, 1173, 1191, 1196, 1214, 1338, 1375
__main__.py	4	4	0%	3–8
config.py	48	7	85%	50, 52–55, 65, 76
TOTAL	2312	237	90%

Tests	Skipped	Failures	Errors	Time
127	0 💤	0 ❌	0 🔥	1m 3s ⏱️