fix(dotcom): lower Zero kill_timeout to Fly's 5m API cap (tldraw#8656)

In order to unblock production deploys, this PR lowers the Zero RM/VS
`kill_timeout` from `10m` back to `5m`. Fly's Machines API rejects
anything over 5 minutes regardless of CPU kind:

```
Error: failed to update machine ...: invalid stop_config.timeout, cannot exceed 5 minutes
```

This contradicts Fly's [graceful VM exits
guide](https://fly.io/blog/graceful-vm-exits-some-dials/), which
suggests up to 24h on dedicated CPU. The 5m cap from the API is the
authoritative limit today. Drain budget is now half what Rocicorp's CZ
uses, but it's the ceiling Fly will accept.

Follow-up to tldraw#8627.

### Change type

- [x] `bugfix`

### Test plan

1. Merge to `production` → `deploy-dotcom.yml` should complete the VS/RM
rolling update without the `invalid stop_config.timeout` error.
2. Verify generated `flyio-view-syncer.toml` and
`flyio-replication-manager.toml` contain `kill_timeout = "5m"` at top
level.

### Code changes

| Section        | LOC change |
| -------------- | ---------- |
| Config/tooling | +4 / -3    |

Author: MitjaBezensek

internal/scripts/deploy-dotcom.ts

@@ -289,8 +289,9 @@ const zeroQueryUrl = `${env.MULTIPLAYER_SERVER.replace(/^ws/, 'http')}/app/zero/
 // Production uses performance (dedicated) CPUs for both RM and VS; staging uses shared.
 // killTimeout: window between SIGTERM and SIGKILL on stop. Lets VS drain client
 // WebSockets and RM flush litestream / release /data handles before being killed.
-// Fly caps it at 5m on shared CPU and 24h on dedicated; production uses 10m to
-// match Rocicorp's CZ default, staging is pinned to the 5m shared-CPU ceiling.
+// Fly's API caps it at 5m regardless of CPU kind (the 24h dedicated-CPU figure
+// from their blog is not actually accepted by the Machines API today — deploys
+// fail with "invalid stop_config.timeout, cannot exceed 5 minutes").
 const zeroVmSizes = {
 	staging: {
 		rm: { cpus: 1, memory: '2gb', cpuKind: 'shared' },
@@ -304,7 +305,7 @@ const zeroVmSizes = {
 		vs: { cpus: 4, memory: '8gb', cpuKind: 'performance' },
 		volumeSize: '8gb',
 		vsMinMachines: 4,
-		killTimeout: '10m',
+		killTimeout: '5m',
 	},
 	preview: { single: { cpus: 2, memory: '2gb' } },
 } as const