Claude review: firmware: qcom: Add a generic PAS service

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

This is the core abstraction layer.

**Race condition in `qcom_pas_ops_register()`:**
```c
void qcom_pas_ops_register(struct qcom_pas_ops *ops)
{
	if (!qcom_pas_is_available())
		smp_store_release(&ops_ptr, ops);
	else
		pr_err("qcom_pas: ops already registered by %s\n",
		       ops_ptr->drv_name);
}
```
This has a TOCTOU race -- two concurrent callers could both see `!qcom_pas_is_available()` as true and both proceed to store. The second store would silently overwrite the first. Consider using `cmpxchg()` instead:
```c
if (cmpxchg(&ops_ptr, NULL, ops) != NULL)
    pr_err(...);
```

**No NULL check on ops callbacks**: The generic functions like `qcom_pas_init_image()` check `!ops_ptr` but never verify that individual callbacks (e.g., `ops_ptr->init_image`) are non-NULL. If a backend registers with a partially populated ops struct, it would crash. This could be a deliberate design choice (all callbacks are mandatory), but it's worth documenting.

**`qcom_pas_metadata_release` has defensive checks the backend doesn't need:**
```c
void qcom_pas_metadata_release(struct qcom_pas_context *ctx)
{
	if (!ops_ptr || !ctx || !ctx->ptr)
		return;
	ops_ptr->metadata_release(ops_ptr->dev, ctx);
}
```
Good -- the outer layer guards mean backends can assume `ctx` and `ctx->ptr` are valid.

**Module with no init function**: `qcom_pas.c` has `MODULE_LICENSE`/`MODULE_DESCRIPTION` but no `module_init`/`module_exit`. It's purely a library module. This works but means the module can be unloaded while backends hold references to the exported symbols. Consider whether this should be a non-module (bool Kconfig) or whether backends should hold a module reference.

---
Generated by Claude Code Patch Reviewer