feat: add session_exists and delete_session to SessionStore

[JiataiWang]

Problem

SessionStore had no way to check whether a session exists by ID, or to
delete a session by reference. Any claw that needed to answer "is this
session already running?" or "clean up this session" had to fall back to
raw filesystem probes against .claw/sessions/<fingerprint>/, bypassing
the workspace-fingerprint namespace and breaking cross-platform portability.

Tracked as ROADMAP #160.

Changes

runtime/src/session_control.rs

• SessionStore::session_exists(session_id: &str) -> bool
  Probes the primary namespace and legacy fallback root with O(1) path
  existence checks — no deserialization, no allocation.

• SessionStore::delete_session(reference: &str) -> Result<(), SessionControlError>
  Accepts any reference the store already resolves: a bare session ID,
  an alias ("latest", "last", "recent"), or a direct file path.
  Delegates to resolve_reference so alias semantics are consistent
  with load_session.

• Module-level free-function wrappers following the existing pattern:
  managed_session_exists / managed_session_exists_for and
  delete_managed_session / delete_managed_session_for.

• Six new tests:

  • session_exists returns true for a persisted session and false for an unknown ID
  • managed_session_exists_for mirrors store behaviour
  • delete_session removes the file and the session no longer appears in list_sessions
  • delete_session via "latest" alias removes the most recent session
  • delete_managed_session_for free-function mirrors store behaviour
  • delete_session returns an error for an unknown reference

Pre-existing clippy errors fixed (unrelated files, caught during CI check):

• runtime/src/bash.rs: format!("{}", now) → format!("{now}")
• runtime/src/plugin_lifecycle.rs: three &&str inefficient .to_string() calls
• commands/src/lib.rs: render_mcp_report_for and render_mcp_report_json_for
  were typed as Result<String/Value, ConfigError> but always returned Ok(...) —
  unwrapped to return the inner type directly; public wrappers and test call-sites updated

Test plan

• cargo fmt --all --check — clean
• cargo clippy --workspace — zero errors
• cargo test --workspace — all pass (flaky repl_executes_python_code and web_fetch are pre-existing environment-dependent tests unrelated to this change)
• New tests cover the happy path, alias resolution, list-after-delete, and the error path for unknown references

🤖 Generated with Claude Code

[code-yeongyu]

This PR has merge conflicts with the latest main. Please rebase on origin/main and force-push. The code changes look good and we're ready to merge once conflicts are resolved.

[JiataiWang]

Rebased on latest origin/main and force-pushed. Conflicts resolved by narrowing the PR scope to just the SessionStore feature (dropped the bundled clippy fixes that were tangential to #160). cargo fmt, cargo clippy --workspace, and cargo test -p runtime are clean. Ready for another look.