feat(tools): add Monitor for polling background tasks

[emal-avala]

Summary

Adds Monitor tool — one call returns status, elapsed time, and the tail of output for a background task.

Why not just TaskGet + TaskOutput? Separate tool calls cost two turns of latency and double the tool-result JSON. Monitor is the combined read, shaped for the polling loop: agent calls it every N seconds while waiting on a build / deploy / long-running command.

Output cap: 8 KiB tail (configurable per-call via tail_bytes, capped). Prevents multi-MB tails from bloating the context mid-poll.

UTF-8 safety: the tail walks past continuation bytes so a multibyte character never gets split at the slice boundary. Prevents the from_utf8_lossy output from showing mojibake when the tail cutoff lands mid-char.

Failure mode: on I/O error (file mid-rotation, transient), returns [output unreadable: \<err\>] instead of erroring — polling should keep working even if a single read fails.

Implementation

• New crates/lib/src/tools/monitor.rs (~200 LOC incl. tests)
• TaskStatus variants mapped to human-readable labels (running / completed / failed: <msg> / killed)
• Elapsed time computed from started_at to finished_at.unwrap_or_else(now)
• Registered in ToolRegistry::with_defaults between TaskOutput and TodoWrite

Test plan

• cargo fmt --all clean
• cargo check passes
• cargo clippy --all-targets -- -D warnings clean
• cargo test -p agent-code-lib --lib tools::monitor — 6 tests (ASCII align, UTF-8 continuation walk, ASCII tail, smaller-than-cap, missing file, multibyte no-split)
• Manual: !cargo build --release & → Monitor tool call with returned task id shows status=running, then status=completed with tail of build output