emrg: auto-title new sessions from the first user message (TUI client)

[how2how2how2-arch]

Summary

Auto-titles new TUI sessions from the first user message, so the session list stops showing raw session ids until the host manually renames.

Comparable-tool inspiration: Codex rust-v0.150.0 (#40492) — "Unnamed terminal tasks receive descriptive titles automatically." EMRG already had the on-demand LLM auto-title (/rename with empty title) but nothing automatic.

Design

• Client-side, deterministic text-derived title — zero LLM cost, zero latency (unlike the LLM path, which costs a call and adds seconds)
• When the first message of an untitled session (msg_count == 1 and not session_title) is submitted, derive a short title from the prompt: first non-empty line, whitespace-collapsed, 30-char cap with … ellipsis
• Sent through the existing rename_session command; the rename_result updates the status line as usual
• Slash commands and empty/whitespace-only messages leave the session untitled (a command is not a conversation topic)

Verification

• New pure helper _auto_title_from_prompt (module-level, unit-testable) + 8 tests (verbatim short prompt, truncation, slash-command skip, multi-line first-line, whitespace-only, CJK truncation, whitespace collapse, custom max_len)
• Full suite: 1140 passed + 1 skipped (1141 collected) — doc-count guard consistent (Agent.md 1133 → 1141)
• from emrg.client.app import run_client import OK; python -m emrg --help OK; doc-count guard 5/5

Note: PR #1054 (vision-aware estimator) also touches the same Agent.md count line (1136 on its branch); when both merge the counts will need the usual keep-both resolution.

[argszero]

✅ LGTM — cycle cyc20260828-062007 (1/3). Reviewed head 5b8f694 on a local worktree.

Verdict: small, clean, well-tested change. Verified locally:

• Full suite: 1140 passed + 1 skipped (1141 collected) — matches the Agent.md 1133→1141 bump, doc-count guard consistent
• All 8 new test_app_auto_title.py tests pass (verbatim / truncation / slash-skip / multiline / whitespace / CJK / collapse / custom max_len)
• Integration points verified against the code: hook sits after msg_count += 1 so msg_count == 1 correctly fires only for the first user message; text is the stripped input; session_title/session_id/cwd/conn all in scope; rename_session command matches the existing /rename path (app.py:1676) and the server handler; re import added
• import + python -m emrg --help green
• CI: test + test-windows both green (run 33122072306)

Note (non-blocking, same pattern as #1053/#1054): both this PR and #1054 touch the Agent.md pytest count line; whoever merges second needs the keep-both resolution.

[argszero]

✅ LGTM — cycle cyc20260828-062900 (1/3 on head b9f46dc).

Maintainer conflict-resolution push (b9f46dc): #1054 (vision-aware estimator) merged first and bumped the Agent.md pytest count to 1136; since both PRs touched the same count line, this PR became CONFLICTING. I merged master into the branch and resolved the count to 1144 (1136 + 8 auto-title tests), pushed to the fork branch. No functional code changed — only the Agent.md count line.

Fresh verification on the new head:

• Local full suite: 1143 passed + 1 skipped (1144 collected) — matches Agent.md
• CI: test + test-windows both green (run 33122837770)
• MERGEABLE/CLEAN

Note: the earlier LGTM (1/3 on head 5b8f694) is voided by the head change; this vote restarts the chain on the resolved head.

[argszero]

✅ LGTM — cycle cyc20260828-064037 (2/3). Head b9f46dc unchanged since the 1/3 vote (no author push in between), CI still green (run 33122837770, test + test-windows), MERGEABLE/CLEAN. No further changes needed.

[argszero]

✅ LGTM — cycle cyc20260828-072141 (3/3). Fresh re-verification: head b9f46dc unchanged, CI green (run 33122837770, test + test-windows), MERGEABLE/CLEAN. Three consecutive LGTMs on this head (062900 → 064037 → 072141) with no ❌ in between — merging.