fix(chain): fail over aborted RPC publish requests

[branarakic]

Summary

• classify raw DOM/undici AbortError and Node ABORT_ERR as retryable chain-RPC transport interruptions
• fail over point reads and rebroadcast the byte-identical signed transaction through the next configured endpoint
• preserve the already-known transaction hash when broadcast endpoints are exhausted, including in the retryable HTTP 503 response

Why

The multi-CG load smoke exposed an ambiguous publish outcome: the chain transaction could be accepted and minted, while the RPC request then aborted and the publisher reported a failure. The raw abort bypassed the existing retry/failover policy because it did not carry one of the previously recognized network error codes.

This change handles that failure at the chain transport seam. It does not re-sign or create a second transaction: failover submits the same signed bytes with the same transaction hash. Deterministic EVM failures remain non-retryable.

This is complementary to #1878. That PR recovers assets that were already stranded after a successful mint; this PR prevents the observed raw-abort path from becoming stranded in the first place and returns the transaction hash if every configured RPC endpoint is unavailable.

Verification

• full dependency and CLI build passed
• full chain unit lane passed
• full CLI unit lane passed
• focused chain failover regressions: 49 passed
• focused CLI transport-status regressions: 9 passed