support session listing and loading

[travisjeffery]

Summary

This PR improves ACP session startup behavior and session selection UX.

It adds strategy-based session loading (latest, prompt, new), supports loading existing sessions when the agent advertises session/list + session/load (codex-acp for example), and has ** Start a new session ** as default.

What Changed

• Added new user option:
  • agent-shell-session-load-strategy (latest default, prompt, new).
• Persisted session capability detection during handshake:
  • tracks support for session/list.
  • tracks support for session/load.
• Updated session initialization flow:
  • if capabilities are present and strategy is not new, attempt session/list then session/load.
  • fallback to session/new on list/load errors or when no session is selected.
• Added session picker helpers:
  • human-readable session labels (title | updatedAt | sessionId).
  • prompt-based session selection.
  • dedicated sorter that pins ** Start a new session ** to the first position in completion display/cycle ordering.
• Refactored session setup internals to reduce duplication:
  • extracted helper for state hydration from response.
  • extracted helper for session-init finalization UI updates.

Tests

Added/updated ERT coverage for:

• list+load preferred path when supported.
• fallback to session/new when session/list fails.
• new strategy bypassing list/load.
• prompt session selection behavior.
• picker sort behavior keeping ** Start a new session ** first.

Targeted test run for these flows passes.

Docs / Follow-up Included in Branch

This branch also includes small documentation updates:

• checkdoc/docstring cleanup in agent-shell.el.
• README variable table entry for agent-shell-session-load-strategy.

Checklist

• I've read the README's Contributing section.
• I've filed a feature request/discussion for a new feature.
• My code follows the project style.
• I've added tests where applicable.
• I've updated documentation where necessary.
• I've run M-x checkdoc and M-x byte-compile-file.
• I've reviewed all code in PR myself and will vouch for its quality.

[xenodium]

Sorry for the delay. We have an existing PR for session loading in #248 (cc @farra). I need to spend some time comparing the two approaches to figure out how to move forward. Relatedly, I'm considering a non-comint buffer alternative which may simplify handshake for us (and possibly loading sessions).

I need to play with both implementations quite a bit. It's gonna take me a little while, sorry.