fix: align per-chip build and P2E runtime paths

[daiyongyuan]

Summary

• build and validate rax-pack with each per-chip compiler
• write and resolve kernel images under output/kernel/<chip>
• keep regression kernel paths consistent with chip-specific outputs
• run the case-local P2E executable with its matching VVAC libstdc++

Validation

• Python syntax parsing passed for all modified files
• git diff --check
• Pebble LeNet completed successfully on P2E