Related: #13374 (TUI compression tip following on resume), #15742 (resolve_resume_session_id compression chain). This PR adds both title lineage resolution via resolve_session_by_title() and compression tip redirection — verify it doesn't conflict with those open PRs.

@alt-glitch thanks for calling this out — I dug into the overlap.

#13374 is definitely the same TUI-side symptom, and it conflicts mechanically with this branch in both merge orders: both touch the session.resume block in tui_gateway/server.py and adjacent protocol tests.

The reason I’d lean toward this shape is that it keeps the resume policy in the shared session layer instead of adding compression-specific logic directly to the TUI gateway. #13374 calls get_compression_tip() from tui_gateway/server.py; this branch routes TUI resume through SessionDB.resolve_resume_session_id(), which is already the shared abstraction used for resume-target normalization. That keeps TUI behavior aligned with the rest of Hermes and gives future resume callers one place to inherit the same semantics.

This branch also covers a second resume issue: title-based resume now goes through resolve_session_by_title(), so /resume Project can resolve to the latest Project #N lineage variant before applying resume-tip resolution. That part is separate from the compression-tip fix and is not covered by #13374.

My read of the relationships is:

• fix(tui): follow compression tips when resuming sessions #13374 is the minimal direct TUI patch for following compression tips.
• fix(state): resume compression continuation tip #15742 is the helper-layer fix that makes resolve_resume_session_id() fully correct for compressed roots that already have messages.
• This PR wires TUI resume into those shared helpers and adds the title-lineage resolution path.

So if maintainers prefer a single landing unit, I’m happy to fold the helper-layer change from #15742 into this branch. Otherwise the clean path seems to be #15742 first, then this PR; #13374’s TUI-specific redirect would be covered by the shared-resolver wiring here.