Stop pci_runtime_pm defaulting to dpv1 devices and surface skipped holds

[timkpaine]

What

dau_utils/pci_runtime_pm.py baked dpv1/Xilinx board identity into a generic host-op utility and defaulted discovery to those patterns whenever the CLI got no --pattern. It also silently skipped writes to absent sysfs paths and still returned success, so a caller could believe PM was held after zero holds landed.

• Remove DEFAULT_DEVICE_PATTERNS; discover_pci_devices now defaults to () (no patterns -> no devices matched). The dpv1 authoritative pattern set stays only in dau-build's platform config.
• main with neither --device nor --pattern now matches nothing and returns 0, without shelling out to lspci. This aligns with dau-build's contract that an empty runtime_pm_patterns produces zero holds (a hold with no --pattern is a successful no-op).
• apply_runtime_pm_writes drops the redundant and write.path.parent.exists() check. main now prints a stderr line per skipped write and exits nonzero with an applied N of M runtime PM writes count when fewer holds are applied than planned.

Notes

dau-build already passes --pattern explicitly per pattern, and the recovery bash scripts use their own lspci discovery, so no consumer relied on the removed default.

Tests cover explicit-pattern discovery, empty discovery, the no-pattern no-op, all-missing and partial (1-of-2) skip diagnostics with nonzero exit, and successful writes, using a tmp_path fake sysfs. pytest dau_utils/tests green (25 passed); ruff check + format clean.