The vite dev <-> vite dev tail handler test was missing the waitForTimeout
parameter on one of its waitFor calls, causing it to use the default (shorter)
timeout which could flake in CI.

…port RPC

Remove the HTTP fallback, eager body buffering, _client GC hack, and WebSocket
special-casing from the dev-registry proxy workers. These workarounds existed
because the workerd debug port would close connections prematurely when the
initial subrequest completed. With the refcounted DebugPortConnectionState fix
in workerd (cloudflare/workerd#6100), connections now survive as long as any
response body or WebSocket is in use.

Key changes:
- Replace hasAssets/entryAddress with defaultEntrypointService and
  userWorkerService fields in WorkerDefinition/RegistryEntry
- Route DOs and named entrypoints through userWorkerService (bypassing
  assets/vite proxy layer) instead of hardcoding core:user: prefix
- Validate required registry fields in resolveTarget() to handle stale entries
- Inline and delete getWorkerdServiceName() helper
- Remove dead entryAddress field and fetchViaHttp code path
- Remove DO WebSocket 501 error and DO body buffering workarounds

…uting

Vite workers with assets were incorrectly routing through assets:rpc-proxy
in the dev registry, but in vite dev mode assets are served by the vite
proxy worker, not workerd's asset pipeline. This caused cross-service
fetch to vite workers with assets to fail (timeout).

Add unsafeOverrideDefaultEntrypoint to miniflare's core options schema so
the vite plugin can explicitly specify which workerd service handles the
default entrypoint. This replaces the brittle inference from
unsafeDirectSockets[].serviceName.

…nd tests

After the debug port RPC refactor, unsafeDirectSockets in the user worker
config, vite plugin, and dev-registry tests no longer serve a functional
purpose — nobody reads the socket URLs. The only remaining use is in
wrangler's ProxyController for InspectorProxyWorker (not touched here).

… cross-worker dev registry

The Reflect.has check already handles keys that exist on the target.
For keys not on the target, the runtime doesn't probe handler properties
through the proxy get trap in a way that requires suppression — confirmed
by tests passing without it.

…ndings instead

Instead of mutating workerOpts with symbol-tagged designators before plugin
processing (requiring every consumer to special-case them), we now let the
plugin system process all bindings normally, then rewrite external service
bindings and tails to point at the dev registry proxy afterward. This follows
the same post-processing pattern already used for assets at line 1673.

Removes kResolvedServiceDesignator, ResolvedServiceDesignator, the zod
validator, and checks in getCustomServiceDesignator, maybeGetCustomServiceService,
and normaliseServiceDesignator.

The proxy's get trap now returns values that are both callable (for
env.SERVICE.method()) and thenable (for await env.SERVICE.property),
matching workerd's JsRpcProperty behavior. Without the .then property,
awaiting a proxy-intercepted property would resolve to the function
itself rather than triggering remote resolution.

Debug port RPC fetchers don't support lifecycle event dispatch
(fetcher.scheduled()), so scheduled events need special handling:

- ExternalServiceProxy: forward scheduled via HTTP to the entry worker's
  /cdn-cgi/handler/scheduled endpoint, which dispatches internally
- RPCProxyWorker (assets proxy): forward scheduled to USER_WORKER via
  Fetcher.scheduled()
- Add service_binding_extra_handlers compat flag to assets proxy service
  config so Fetcher.scheduled() is available on the USER_WORKER binding

Refactors the  to be more generic, efficient,
and readable.

- Introduces a private  method that handles
  resolving and connecting to the remote service.
- Caches the remote fetcher promise for 1s to avoid re-connecting
  on every single RPC call or property access.
- Simplifies the Proxy  trap by using the new helper, removing
  duplicated code and making the logic much clearer.

…, thenable DO RPC

…N, fix stale cache clear

… aggregation

…ocket

On Windows, the internal Cap'n Proto service-to-service forwarding from
the entry worker to the dev-registry-proxy service can break with
WSARecv error 64 (ERROR_NETNAME_DELETED), causing the registry push to
silently fail and leaving the proxy worker with an empty registry.

Give the dev-registry-proxy service its own HTTP socket (SOCKET_DEV_REGISTRY)
and have #pushRegistryUpdate POST directly to it, bypassing the entry worker
entirely. Also incorporates review fixes: reset previousJSON on restart,
guard generated identifiers, check dispose signal on retry, read fresh
registry on retry, shallow-copy workerOpts before mutation, use real entry
port for loopbackAddress, and guard malformed loopbackAddress parsing.