Enable CNTVCT on cloned vCPUs

[doanbaotrung]

Worker vCPUs created for clone threads did not get CNTKCTL_EL1 configured, even though the initial vCPU enables EL0 access to the architected counter.

Set CNTKCTL_EL1 during cloned vCPU setup so CNTVCT_EL0 reads work from guest threads as they do on the main vCPU. Add a pthread regression test that reads CNTVCT_EL0 from both main and worker vCPUs.

Fix #229

---

Summary by cubic

Enable EL0 reads of the architected timers on all vCPUs by using a single CNTKCTL_EL1 mask in bootstrap and both cloned-vCPU paths (CLONE_THREAD and CLONE_VM). Restores CNTVCT_EL0/CNTPCT_EL0 for guest threads and adds a regression test. Fixes #229.

• Bug Fixes
  • Define CNTKCTL_EL1_EL0_TIMER_EN in hvutil.h; apply it in bootstrap, thread_create_and_run, and vm_clone_thread_run to enable CNTVCT_EL0 and CNTPCT_EL0 at EL0.
  • Add tests/test-cntvct-thread.c; verify CNTVCT_EL0 on the main vCPU and a CLONE_THREAD worker. Add to test matrix and update baselines (elfuse-aarch64 238, qemu-aarch64 218).

^{Written for commit 37c3c53. Summary will update on new commits.}

[doanbaotrung]

Implemented the reviewed fixes in third_party/elfuse:

• Added shared CNTKCTL_EL1_EL0_TIMER_EN in hvutil.h.
• Replaced all three 0x3ULL CNTKCTL sites in bootstrap.c and forkipc.c.
• Expanded test-cntvct-thread.c to:
  • read both CNTVCT_EL0 and CNTPCT_EL0
  • cover pthread_create / CLONE_THREAD
  • cover raw clone(CLONE_VM | SIGCHLD) for vm_clone_thread_run
  • fail loudly on non-aarch64 builds with #error
  • document the current “trapped MRS returns 0” coupling

[jserv]

Thank @doanbaotrung for contributing!