Claude review: soc/tegra: pmc: Move legacy code behind CONFIG_ARM guard

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

Patch Review

Moves the remaining legacy functions (`tegra_pmc_cpu_is_powered`, `tegra_pmc_cpu_power_on`, `tegra_pmc_cpu_remove_clamping`, `tegra_pmc_core_domain_state_synced`, and suspend mode functions) behind `#if defined(CONFIG_ARM)` in both `pmc.c` and `pmc.h`.

The `early_pmc` static variable itself is also guarded by `CONFIG_ARM`, and `#if defined(CONFIG_ARM)` blocks are added around the early_pmc usage in `tegra_pmc_probe()`.

The `tegra_pmc_early_init()` function and its `early_initcall()` are also moved behind the CONFIG_ARM guard. This means on arm64, the early init code won't run at all -- which is correct since the dynamic allocation in probe handles everything.

In `pmc.h`, the header is restructured to separate the instance-aware API (guarded by `CONFIG_SOC_TEGRA_PMC`) from the legacy 32-bit-only API (guarded by `CONFIG_ARM`). The suspend mode stubs are now under `CONFIG_ARM && !(CONFIG_SOC_TEGRA_PMC && CONFIG_PM_SLEEP)`. A `COMPILE_TEST` stub for `tegra_pmc_core_domain_state_synced()` is provided for the `!CONFIG_ARM` case.

The `#if defined(CONFIG_ARM)` blocks inside `tegra_pmc_probe()` are a bit ugly (in-function `#ifdef`), but acceptable as a pragmatic solution. An alternative would be a helper function, but the current approach is clear enough.

The cover letter mentions "Changes in v2: fix build issue for ARM && !PM_SLEEP" which only appears in this patch's changelog but is formatted as if it's a per-patch note rather than a series-level note. Minor formatting nit.

No functional issues.

---
Generated by Claude Code Patch Reviewer