Stop leaking goroutines when multiple submissions fail together

[damilolaedwards]

Summary

• The manager for a wallet's batch submission drains only the first error from a size-1 error channel, then stops listening. Every other failing sub-goroutine was doing an unconditional blocking send on that same channel, so once the manager stopped draining, they blocked forever with no way to release their semaphore slot or run their own cleanup.
• Changed the send to a non-blocking best-effort select, so a sub-goroutine that fails after the manager has already moved on drops its error instead of hanging.

Test plan

• Added a test that mirrors the manager/sub-goroutine channel structure with several sub-goroutines failing at once, and checks no goroutines are left running after the manager returns.
• Verified the test fails against the old blocking send and passes with the fix, under -race.