Reorder core tick pipeline and make driver I/O progress per-port

[Copilot]

This PR updates cycle behavior to avoid barrier-induced stalls: the core now receives before executing and sends after execution, and driver FeedIn/Collect now advance independently per port instead of waiting for full-round synchronization. It also aligns task validation and tests with the new progress model.

• Core tick sequencing

  • Reordered core.Tick() from send/execute-style flow to:
    • doRecv() -> runProgram() -> doSend()
  • Improves forward progress when inputs arrive asynchronously across ports.

• Driver FeedIn/Collect semantics

  • Replaced global round tracking with per-port progress (portRounds) in both feedInTask and collectTask.
  • Feed-in/collect loops now opportunistically process whichever ports are ready in the current cycle.
  • Task completion is determined by all ports reaching rounds, not by lockstep round barriers.

• Layout safety and failure behavior

  • Added/used validateTaskLayout(...) to enforce valid (dataLen, stride, portCount) combinations up front.
  • Out-of-range index paths in active processing are treated as invariant violations (panic), preventing silent drops/stalls from inconsistent task configuration.

• Test updates

  • Updated internal driver tests to the new task fields and semantics (rounds + portRounds).
  • Added/updated opportunistic behavior assertions for non-uniform port readiness in both FeedIn and Collect paths.
  • Removed obsolete barrier-era helper usage.

func (c *Core) Tick() (madeProgress bool) {
	madeProgress = c.doRecv() || madeProgress
	madeProgress = c.runProgram() || madeProgress
	madeProgress = c.doSend() || madeProgress
	return madeProgress
}

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.