fix(www): keep committed agent-skills index when fetch fails outside production (#50556)

## I have read the
[CONTRIBUTING.md](https://github.com/supabase/supabase/blob/master/CONTRIBUTING.md)
file.

YES

## What kind of change does this PR introduce?

Bug fix (build resilience)

## What is the current behavior?

`apps/www/scripts/fetchAgentSkills.mjs` runs as part of `content:build`
and fails the whole `www` build (and `pnpm dev:www`) whenever the GitHub
API call fails. #50106 added `AGENT_SKILLS_GITHUB_TOKEN` to mitigate
rate limits on Vercel, but that does not cover local runs or builds
where the env var is not available (e.g. fork PRs).

Example from a local `pnpm dev:www` hitting the unauthenticated rate
limit:

```
www:dev: Error: GET https://api.github.com/repos/supabase/agent-skills/releases/latest → 403
www:dev: at fetchJson (file:///.../apps/www/scripts/fetchAgentSkills.mjs:38:22)
www:dev: at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
www:dev: at async main (file:///.../apps/www/scripts/fetchAgentSkills.mjs:53:19)
```

## What is the new behavior?

`public/.well-known/agent-skills/index.json` is already committed to the
repo, so when the fetch fails and `VERCEL_ENV` is not `production`, the
script logs the error, keeps the committed file, and exits 0:

```
www:dev: Error: GET https://api.github.com/repos/supabase/agent-skills/releases/latest → 403
www:dev: ...
www:dev: Fetch failed — keeping committed public/.well-known/agent-skills/index.json
```

Production builds still fail loudly so a stale skills list is never
silently shipped.

Verified locally by forcing a 401 with a bad token:

- `VERCEL_ENV=preview` exits 0 and keeps the committed `index.json`
- `VERCEL_ENV=production` exits 1

## Additional context

Follow-up to #50106.

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **Bug Fixes**
  * Improved handling of skill data fetch failures outside production.
* Preserves previously available skill data when a fetch fails and a
committed fallback is available.
  * Continues to report failures when no fallback data exists.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

apps/www/scripts/fetchAgentSkills.mjs

@@ -10,9 +10,12 @@
  * Spec: https://github.com/agentskills/agentskills/pull/254
  * Uses AGENT_SKILLS_GITHUB_TOKEN if set to avoid GitHub's unauthenticated
  * rate limit (60 req/hr per IP, shared across Vercel build machines).
+ *
+ * If the fetch fails outside production, the committed index.json is kept so
+ * preview and local builds don't break on GitHub rate limits.
  */
 
-import { promises as fs } from 'node:fs'
+import { existsSync, promises as fs } from 'node:fs'
 import { dirname, join } from 'node:path'
 import { fileURLToPath } from 'node:url'
 
@@ -50,5 +53,12 @@ async function main() {
 
 main().catch((err) => {
   console.error(err)
+  const canFallBack = process.env.VERCEL_ENV !== 'production'
+  const hasPreviousWrite = existsSync(join(OUT_DIR, 'index.json'))
+
+  if (canFallBack && hasPreviousWrite) {
+    console.warn('Done — keeping committed public/.well-known/agent-skills/index.json')
+    process.exit(0)
+  }
   process.exit(1)
 })