Claude review: gpu: nova-core: convert to new dma::Coherent API

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

Patch Review

This is the final cleanup patch converting all remaining `CoherentAllocation` usage.

**`from_data` safety scope**:
```rust
unsafe { dma_obj.as_mut()[..data.len()].copy_from_slice(data) };
```
The `unsafe` block wraps the entire expression, but only `as_mut()` is the unsafe operation. The slice indexing and `copy_from_slice` are safe. This is a minor style issue — narrower unsafe blocks are preferred in kernel Rust.

**`falcon.rs` change**:
```rust
// Before:
fw.dma_handle_with_offset(load_offsets.src_start.into_safe_cast())?
// After:
fw.dma_handle() + DmaAddress::from(load_offsets.src_start)
```
The original used `dma_handle_with_offset` which performed bounds checking (returning `EINVAL` if out of range). The new code does raw arithmetic without bounds checking. This is a subtle semantic change — if `src_start` were ever invalid, the old code would return an error while the new code would silently produce a wrong DMA address. Worth confirming this is intentional.

**`PteArray` adds `FromBytes`**:
```rust
/// SAFETY: arrays of `u64` implement `FromBytes` and we are but a wrapper around one.
unsafe impl<const NUM_ENTRIES: usize> FromBytes for PteArray<NUM_ENTRIES> {}
```
This is correct — `[u64; N]` is trivially `FromBytes` and the `#[repr(C)]` wrapper preserves this.

**`cmdq.rs` simplifications**: The replacements of `start_ptr()` → `as_ptr()` and `start_ptr_mut()` → `as_mut_ptr()` throughout are mechanical and correct. The removal of "CoherentAllocation contains exactly one object" from safety comments is appropriate since `Coherent<GspMem>` is inherently a single object.

---
Generated by Claude Code Patch Reviewer