Add regression test BSP, build scripts, and CI for RV32/RV64

[akifejaz]

Adds QEMU-virt BSP and regression test infrastructure for RISC-V 32/64-bit ports.

• QEMU-virt board support package (CLINT timer, PLIC, UART, trap handler, linker script)
• Build/test scripts (run.sh, build_tx_riscv.sh, test_tx_riscv.sh) running 5 configs × 2 architectures (950 tests)
• CI integration via regression_test.yml using xPack toolchain + QEMU
• Fix 64-bit portability issues in regression tests: block/byte pool sizes, byte pool search pointer offset, timing tolerance, -Wconversion in trace code

All regression tests passes for RV32/RV64, please see the pipeline status here

[fdesbiens]

plic_claim() can return any value 0–1023, but callbacks[] only has MAX_CALLBACK_NUM=128 entries. There is no bounds check before callbacks[irqno] is accessed.

 // Fix: add bounds check at top of plic_irq_intr()
 int irqno = plic_claim();
 if (irqno <= 0 || irqno >= MAX_CALLBACK_NUM) {
     plic_complete(irqno);
     return 0;  // spurious or out-of-range, not an error
 }

Note: checking irqno == 0 also handles the PLIC spurious-interrupt case (irqno=0 means "no interrupt").

[fdesbiens]

(1 << irqno) is UB when irqno >= 32 (int overflow). The PLIC enable registers are also word-arrays, so only IRQs 0–31 are correctly addressable.

Fix: use (1U << (irqno % 32)) with the correct word offset, or add a guard that asserts/returns if irqno >= 32.

[fdesbiens]

@akifejaz I reviewed the PR and everything looks good. Please add the two small comments above on the PLIC, and we should be good to go!