Cp pci domain fix

[manoj-freyr]

Fix PEQT segfault and kernel_driver detection on multi-domain PCIe systems

PEQT, SMQT, and PESM matched PCI devices to GPUs using only bus/dev/func
(location_id) and ignored the PCIe domain. On systems with multiple domains
(e.g. 0000: and 0001:), unrelated devices can share the same location_id
— a GPU at 0000:05:00.0 and a bridge at 0001:05:00.0 both map to 0x0500.
Those modules then ran PCIe capability queries on the wrong device; libpci
state (notably dev->access) was corrupted and get_kernel_driver() segfaulted
dereferencing it.

Use domlocation2gpu(domain, location_id) so only the intended GPU BDF is
matched.

Clanmup and Hardening changes:

• get_kernel_driver: read /sys/bus/pci/devices//driver instead of
  libpci dev->access/readlink, avoiding bad access pointers while still
  reporting amdgpu.
• get_pwr_budgeting: skip when config access is unavailable; continue the
  DSR loop on failed/empty reads instead of aborting mid-query.
• PEQT dynamic power-budget capabilities: validate map lookups before use.