[tim-smart]

Chat status messages stopped moving past "Starting" after #179.

#179 also moved the effect snapshot forward on eff-698-cloudflare-cluster. In that snapshot the Cloudflare entity runtime runs handlers through RpcServer, and RpcServer gives each request its own Scope, closed when the handler returns. Before, handlers saw the entity's scope. Alchemy's Postgres layer keeps one pool per Scope in context, and an effect/Pool whose scope is closed answers every get with an interrupt.

The session entity forks a turn into entityScope from the Receive handler, but the forked fiber still carried that request's Scope. Once Receive returned, every Postgres query from the turn was interrupted. The status watcher reads the kill switch on each tick. ChatKillSwitch.get catches failures but not interrupts, so the first tick killed the watcher and the status stayed at "Starting". saveIndex logged and went on, which is why production's chat_sessions rows still say running for turns that have replied.

This adds forkInEntity in Chat/Session.ts: it forks into the entity scope and also provides that scope as the fiber's Scope, as handlers had before. It replaces all four forkIn(..., entityScope) calls.

The new test in Session.test.ts uses a kill switch that interrupts once the caller's scope is closed, the way the production pool does. It fails without the fix ("Timed out waiting for a Working status") and passes with it.

PullRequestReview/RunEntity.ts also forks into its entity scope. I haven't checked whether any of that work queries Postgres after its request ends.

Closes EFF-3451