[Sync] Correct inter-core sync lowering semantics for A3/A5

[TaoTao-real]

Summary

This PR only contains inter-core sync support changes:

• Keep IR shape unchanged: pto.sync.set/pto.sync.wait
• Align lowering semantics with ISA differences:
  • A3
    • sync.set -> ffts_cross_core_sync(pipe, getFFTSMsg(FFTS_MODE_VAL, event_id))
    • sync.wait -> wait_flag_dev(event_id)
  • A5
    • sync.set -> set_intra_block(pipe, event_id)
    • sync.wait -> wait_intra_block(pipe, event_id)
• Add shared lowering utility for inter-core sync call selection/argument assembly in PTOToEmitC
• Add pto.set_ffts EmitC lowering to set_ffts_base_addr(...)
• Add A3 compile-time hard check: using inter-core sync ops without pto.set_ffts in the same function now fails
• Add Python helper wrappers: sync_set, sync_wait, set_ffts
• Add Python regression samples:
  • test_intercore_sync_a3.py
  • test_intercore_sync_a5.py
  • test_intercore_sync_a3_missing_setffts.py
• Extend runop.sh with arch-aware routing + structural assertions for the new samples

Scope guard

Diff vs main contains only 6 files:

• lib/PTO/Transforms/PTOToEmitC.cpp
• python/pto/dialects/pto.py
• test/samples/runop.sh
• test/samples/Sync/test_intercore_sync_a3.py
• test/samples/Sync/test_intercore_sync_a5.py
• test/samples/Sync/test_intercore_sync_a3_missing_setffts.py

Validation

Local validation passed:

• ninja -C build ptoas install
• bash test/samples/runop.sh -t Sync (A3 default)
• PTOAS_FLAGS='--pto-arch=a5 --enable-insert-sync' bash test/samples/runop.sh -t Sync

Both A3/A5 new regression assertions are hit as expected, and the A3 missing-set_ffts case is XFAIL with the expected diagnostic.