ops: error tracking and session diagnostics

[joelpeace48-cell]

Problem

Client-side errors are invisible. A wallet connection that fails for a specific
browser or extension version, or a dashboard that throws on an edge-case data
shape, will simply be experienced as "the site is broken" with no signal
reaching anyone who could fix it.

What to do

• Integrate Sentry or an equivalent, capturing client and server errors with
  source maps uploaded at build time — a minified stack trace is not
  actionable.
• Wire it into the error.tsx and global-error.tsx boundaries.
• Scrub aggressively before sending: no wallet addresses, no request bodies, no
  anything that identifies a user. This is a privacy-forward product and the
  error tracker must not become the leak.
• Tag events with release version and commit SHA so a regression is traceable
  to a deploy.
• Alert on error-rate spikes rather than individual events.

Acceptance criteria

• Client and server errors captured with source maps
• Wired into both error boundaries
• PII and wallet data scrubbed, verified by a test
• Releases tagged with commit SHA
• Rate-based alerting configured

Notes

Write the scrubbing test as an assertion that a known address string never
appears in a captured payload. It is the only form of this that stays true as
the codebase changes.