fix: add structured request context to all API error logs

[YizukiAme]

Summary

Replace generic log.error('Error:', error) messages across 23 API routes with structured, context-rich error logs for faster incident triage.

Closes #323

Changes

Pattern applied uniformly

// Before
log.error('Error:', error);

// After
log.error(`TTS generation failed [provider=${ttsProviderId ?? 'unknown'}, voice=${ttsVoice ?? 'unknown'}]:`, error);

What's included

• Hoist key request params (let x: T | undefined) above try blocks so they're accessible in catch
• Add createLogger imports to 4 routes that previously had no logging (classroom, classroom-media, generate-classroom/*)
• Truncate user-controlled strings (.substring(0, 60) / .substring(0, 100)) to prevent log flooding
• No behavioral changes — apiError() responses are untouched; only log messages improved

Affected routes (23 files)

Category	Routes
Generation	generate/image, generate/video, generate/tts, generate/agent-profiles, generate/scene-actions, generate/scene-content, generate/scene-outlines-stream
Classroom	classroom, classroom-media, generate-classroom, generate-classroom/[jobId]
Chat	chat, pbl/chat
Verification	verify-model, verify-image-provider, verify-video-provider, verify-pdf-provider
Other	azure-voices, parse-pdf, proxy-media, quiz-grade, transcription, web-search

Context logged per route

Each error log now includes the most relevant request parameters for that endpoint:

• Provider/model identifiers — for provider-related routes
• Input summaries — truncated requirement text, query strings, question snippets
• Resource identifiers — audioId, classroomId, jobId, stageId

Testing

• Purely observability changes — no runtime behavior modified
• All existing error responses (apiError(...)) unchanged
• Verified no new imports break the build

---

As the changes are largely mechanical, all code edits in this PR were made by Claude Opus 4.6, but have been manually audited and tested.

[YizukiAme]

Test Results — Before / After Log Comparison

Tested by sending deliberately invalid credentials/inputs to 5 API routes, capturing server-side log.error() output from the dev server.

Before (original generic messages)

[ERROR] [Verify Model] API test error: AI_APICallError: Incorrect API key provided ...
[ERROR] [Azure Voices] API error: TypeError: fetch failed
[ERROR] [WebSearch] [WebSearch] Error: Error: Tavily API error (401): { "error": "Unauthorized" }
[ERROR] [PBL Chat] Error: TypeError: Cannot read properties of undefined ...
[ERROR] [Transcription] Transcription error: TypeError: Content-Type was not one of ...

After (structured context logs from this PR)

[2026-03-30T21:08:23Z] [ERROR] [Verify Model] Model verification failed [model="gpt-999-turbo-fake"]: AI_APICallError: Incorrect API key provided: sk-INVAL**TING.
[2026-03-30T21:08:24Z] [ERROR] [Azure Voices] Azure voices fetch failed [baseUrl="https://bad-endpoint.invalid/cognitiveservices/v1"]: TypeError: fetch failed
[2026-03-30T21:08:26Z] [ERROR] [WebSearch] Web search failed [query="test query for error logging demonstration"]: Error: Tavily API error (401): { "error": "Unauthorized: missing or invalid API key." }
[2026-03-30T21:08:27Z] [ERROR] [PBL Chat] PBL chat failed [agent="TestAgent", type=question]: TypeError: Cannot read properties of undefined (reading 'length')
[2026-03-30T21:08:28Z] [ERROR] [Transcription] Transcription failed [provider=unknown, model=default]: TypeError: Content-Type was not one of "multipart/form-data" or "application/x-www-form-urlencoded".

Key improvements visible in logs

Route	Before	After
verify-model	No model name	model="gpt-999-turbo-fake"
azure-voices	No URL	baseUrl="https://bad-endpoint.invalid/..."
web-search	Duplicate prefix, no query	query="test query for error logging..."
pbl/chat	Error: only	agent="TestAgent", type=question
transcription	Transcription error	provider=unknown, model=default

Note: provider=unknown in Transcription is correct — the test intentionally sent JSON instead of multipart/form-data, so req.formData() threw before resolvedProviderId was ever assigned.