fix(dashboard): scope PTY attach token by resume session

[thelonewander3r]

Summary

• Scope the dashboard Chat keep-alive PTY attach token by resume session id and profile in localStorage, instead of one shared hermes.pty.token.chat key for the whole tab.
• Fixes sidebar session switching after New chat: clicking an older session now loads that session's transcript instead of reattaching the wrong living PTY.
• Complements backend attach-key scoping from fix(dashboard): scope chat attach tokens by session #60745; this is the missing frontend half of the regression reported in Dashboard sidebar session resume does not display conversation history (v0.18.x regression) #64425 / **Web dashboard: clicking a previous session opens a blank new session (/api/pty ?attach= reattach discards the ?resume= spawn env)** #63701 / Windows Desktop can timeout when packaged Electron readiness parser is older than Python backend #60772.
• Related open PR fix(dashboard): scope PTY attach token per resume session to prevent wrong-session reattach #61045 overlaps on the frontend token scope idea; this change is intentionally narrow (frontend-only) and verified against current main.

Test plan

• On WSL dashboard (hermes dashboard → /chat): create chat A with history → New chat B → New chat C → click A in the sidebar → A's transcript loads
• Hard-refresh after deploying the rebuilt web_dist so the new bundle is used
• Refresh while staying on the same ?resume= session still reattaches keep-alive (same scoped token)
• Profile-scoped chat (?profile=) keeps a separate attach token from the default profile
• Existing /api/pty keepalive tests still pass: scripts/run_tests.sh tests/test_pty_keepalive_ws.py -q

[alt-glitch]

This was generated by AI during triage.

Related to #64154, #64163, and #61045. This is the clean profile+resume frontend scope; the open alternatives retain broader lifecycle work or resume-only semantics.

[thelonewander3r]

After re-review against current main and #60745: closing this PR.

Backend #60745 already scopes the keep-alive PTY registry by canonical resume + profile, with regression coverage for session switches and parent→descendant reattach (test_attach_token_reuses_canonical_resume). Frontend scoping by the raw resumeParam localStorage key is redundant on that base and can mint distinct attach tokens for parent vs child before the server canonicalizes them, which risks duplicate PTYs for one conversation.

Our local A→B→C→A “fix” was likely from rebuilding web_dist / hard-refresh rather than this hunk being the real attach-key fix. Also dropping the incorrect #60772 reference — that issue is unrelated (Windows Desktop readiness parser).

If #64425 / #63701 still reproduce on current main without this change, they need a different root-cause investigation (e.g. descendant/subagent walk or history render), not another attach-token layer. Leaving those issues for verification rather than claiming Fixes here.

Sorry for the noise — thanks for the triage note linking #61045 / related work.

[thelonewander3r]

Closing as superseded by #60745; see comment above.