fix(core): broadcast wellknown updates globally and bound HTTP timeouts

[kitlangton]

Why

WellKnown is a global service shared across all project Locations, and #53667 (f108fa4928) now persists its manifest and remote config in KV across transient refresh failures. Two gaps remain:

1. WellKnown.refresh, WellKnown.add, and WellKnown.remove publish WellKnown.Event.Updated without { global: true }, so only the Location that triggered the update reloads while other open Locations remain on the previous manifest.
2. WellKnown.inspect and WellKnown.resolveEntry have no request deadline, so a hung socket during a VPN/WARP cutover holds WellKnown's lock and Config's reload lock until the OS socket timeout (up to 6 minutes) before falling back to the cached KV snapshot.

What Changes

• Publish WellKnown.Event.Updated with { global: true } in refresh, add, and remove so every open Location reloads when a well-known source changes.
• Add a 15-second timeout (Effect.timeout("15 seconds")) to WellKnown.inspect and WellKnown.resolveEntry so hung connections fall back promptly to the cached KV config.

Verification

cd packages/core
bun run test test/wellknown.test.ts
bun typecheck