Fix Wiki Writer timeouts misreported as transient API errors

[Copilot]

The agent was loading 3000+ lines of source into context then generating all wiki pages in a single push-wiki call, exceeding output token/timeout limits. The API returned a timeout; the CLI logged it as "transient" and retried into the same wall repeatedly.

Changes

Mandatory batching constraints (3d)

New hard-constraint block the agent must read before generating anything:

• Max 4 pages per push-wiki call; process sequentially
• No sub-agents — generation must happen in the main conversation loop
• Per-page limit: 3 KB; per-call payload limit: 30 KB
• On API error: retry up to 2× halving batch size each time (min 1 page/call); single-page failures are unrecoverable

Large-file context limits (3d)

For source files >500 lines, forbids reading the full file. Instead:

grep -n "FunctionName\|ClassName" src/foo.ts | head -20  # locate relevant lines
head -n 150 src/foo.ts | tail -50                        # extract that region

Batched push-wiki (3f)

Replaces the previous "one giant JSON object" instruction with explicit sequential batching — sidebar included only in the final batch, payload size estimated before each call, and consistent retry semantics aligned with the 3d constraints.

Error clarity

Retry guidance explicitly names timeouts as the most likely cause of API errors during generation, so the agent (and operators reading logs) know what's actually happening rather than chasing phantom network issues.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.