relay: events with embedded NUL (\u0000) reach Postgres and fail with UTF8 0x00 internal error instead of clean rejection

**What happens**

Production relay pods (bb-public cell) intermittently log:

```
Internal error: error: database error: database error: error returned from database: invalid byte sequence for encoding "UTF8": 0x00 at line 1870
```

27 occurrences over the last 14 days (first sighting 2026-07-22, ~5/day since, most recent 2026-07-27 12:14Z), across many different pods/hosts, often in pairs seconds apart — consistent with a client retrying after a 500. Datadog: https://app.datadoghq.com/logs?query=kube_cluster_name%3Aproduction-bke-bb-public-infra-cell-01-us-west-2%20kube_namespace%3Abuzz%20%22invalid%20byte%20sequence%22

**Likely root cause**

JSON permits `\u0000` escapes in strings, but Postgres `text`/`jsonb` columns cannot store NUL bytes. An event whose content (or a tag value) contains an embedded NUL passes JSON parsing and signature checks, then blows up at insert time as a Postgres encoding error, which surfaces as `Internal error` (500 / generic failure) to the client instead of a clean `invalid:` rejection.

**Expected**

Events containing NUL bytes in content or tag values should be rejected at validation time (`invalid: content contains disallowed control characters` or similar), before any DB write — not surfaced as a database internal error.

**Notes**

- Blast radius today is one client request per hit (no panic/restart, the transaction just fails), but each hit is a 500 to a real user and an error-stream pollution source during launch monitoring.
- Worth checking all ingest text paths (event content, tags, profile fields, workflow definitions) for the same gap, since the error doesn't identify which statement fired.