fix(cli): retry registry requests and fall back to IPv4

[dennytosp]

Fixes #39

Problem

npx reacticx list / reacticx add <component> fail on some networks with a raw undici error:

[TypeError: fetch failed] {
  [cause]: ConnectTimeoutError: Connect Timeout Error
      (attempted addresses: 2606:4700:310c::ac42:2ca8:443, timeout: 10000ms)
    code: 'UND_ERR_CONNECT_TIMEOUT'
}

Cause

getRegistry and getComponentCode each called fetch exactly once, with no timeout, no retry, and no error handling beyond response.ok. On a dual-stack network that advertises an AAAA record for the registry host but cannot actually route IPv6, the first connect hangs until undici gives up — and there is nothing to recover it. The reporter also had no way to raise the timeout, and the thrown error gave no hint about what to try.

Fix

Both requests now go through a fetchWithRetry helper that:

• retries up to 3 times with an incremental backoff — fetch only rejects on transport-level failures (an HTTP error status resolves normally), so anything thrown is worth another attempt;
• switches the DNS resolver to ipv4first after a connect-level failure (UND_ERR_CONNECT_TIMEOUT, EHOSTUNREACH, ENETUNREACH, ETIMEDOUT), so a broken IPv6 route recovers on the next attempt. This is only applied as a fallback, so IPv6-only networks are unaffected;
• applies an explicit 30s timeout, overridable with REACTICX_FETCH_TIMEOUT=<milliseconds>;
• explains what to try when every attempt fails, while preserving the original error as cause so the underlying code is still visible.

Failed responses now also report their HTTP status (Failed to fetch registry (404 Not Found)) instead of a bare message.

Verification

Against the live registry — list, list -c organisms, and add all behave as before.

With a stubbed fetch, exercising each path:

scenario	result
2 connect timeouts then success	recovers on attempt 3
connect timeout on every attempt	3 attempts, then the guidance error, cause = UND_ERR_CONNECT_TIMEOUT
connect-level failure	resolver flips to ipv4first
non-connect failure (ECONNRESET)	retries, resolver left at verbatim
HTTP 404	1 attempt, no retry
REACTICX_FETCH_TIMEOUT=1234	honoured, and quoted back in the error

tsc --noEmit and tsup both pass. No new dependencies — node:dns only.