fix(cli): retry processor upload on ECONNRESET and other transient network errors

[jupiterv2]

Problem

sentio upload fails intermittently at the final step that PUTs the packed
processor (dist/lib.js) to the signed storage URL, with:

FetchError: request to https://storage.googleapis.com/sentio-processors/upload/... failed, reason: socket hang up
  code: 'ECONNRESET'

Build / codegen / packaging all succeed and the signed URL is obtained fine —
only the upload PUT (via node-fetch) gets reset. The same PUT to the same GCS
host succeeds reliably via curl and via Node's built-in fetch (undici), so
the reset is a transient connection issue that node-fetch 2.x surfaces and
does not transparently retry.

Root cause

The retry guard in tryUploading only matched EPIPE:

if (e?.constructor.name === 'FetchError' && e.type === 'system' && e.code === 'EPIPE') {

ECONNRESET (the common one against GCS) fell through to the else branch and
aborted the whole upload after a single attempt — no retry — even though the
triedCount/backoff machinery was already in place for up to 5 tries.

Fix

Widen the retry whitelist to the common transient socket errnos
(ECONNRESET, ETIMEDOUT, ECONNREFUSED, EAI_AGAIN) in addition to EPIPE.
No behavior change for non-transient errors (they still surface immediately).

Test

Reproduced the ECONNRESET reliably (intermittent) with node-fetch 2.7.0
PUT-ing ~5MB to the GCS bucket host; undici and curl never reset. After the
change, a reset triggers the existing 1s-delay retry loop instead of aborting,
which is what lets the upload succeed on a subsequent attempt.