ctx.flybot host API: change-seq reset, MCP client timeouts, reserved names (PR #2 review) #18

Context

The PR #2 review listed several P2 robustness gaps in the host ctx.flybot service and the social MCP client. They are spot-checked as still present at 7b806c6 (for example, changeSeq is still an in-process counter, src/host/hooks.js#L86-L89). Related to #4.

Original feedback (@hsliuustc0106, permalink, P2)

• State/persistence: "changeSeq resets per process while the browser watermark never regresses → field events silently stop after every host restart". In addition, the corrupt-fields reset in store.js does not set migrated, and change.value gives listeners the stored object, so a listener that mutates it diverges memory from disk.
• MCP client: "no timeout/AbortSignal anywhere (a hung server stalls polling, transport, and approvals.decide indefinitely), and it advertises text/event-stream in accept but would crash parsing an SSE response."
• API surface: "api.handle silently accepts built-in names (approvals, decide, …) that the endpoint switch shadows — reject them. flybot.scope('shared') is allowed and collides with the unscoped namespace."

Proposed fix

• Add a per-process epoch to change events (or persist changeSeq), and have the client reset its watermark when the epoch changes.
• Set migrated on reset. Deep-clone or freeze change.value.
• Add a timeout or AbortSignal to every MCP call. Either handle SSE or stop advertising it.
• Reject reserved endpoint names and the shared scope.

Acceptance criteria

• Field events keep flowing to an open browser after a host restart.
• A hung MCP server causes a timeout error, not a stall.
• handle('approvals') and scope('shared') throw.
• Tests cover each case.

Refs: PR #2 (merged), umbrella #3. Verified against Birfy/FlyBot flybot-plugins tip 7b806c6.