lib: buffer - Annotate w_buffer with sized_by(alloc)

[LaptopsPlural]

Summary

real_buffer.w_buffer is the private writable allocation whose capacity is tracked by alloc. Annotate that relationship with sized_by(alloc) when the compiler provides the attribute (no-op otherwise), and in buffer_alloc() assign the new alloc before publishing the new pointer so the bound remains consistent while the buffer grows.

No change to public buffer_t layout, writable_size/NUL reservation, or unsafe space helpers.

AI-assisted disclosure

This change was prepared with AI assistance. I reviewed the diff for correctness and scope before opening this PR.

[sirainen]

Probably not useful to check for __sized_by__?

[LaptopsPlural]

Agreed — dropped the __sized_by__ spelling check. We now only gate on __has_attribute(sized_by) (empty macro otherwise).

[sirainen]

Should we add this to our malloc() variants?

[LaptopsPlural]

Thanks for the review.

On malloc variants: I think yes as a follow-up, not in this PR. This change is intentionally one concern — private w_buffer↔alloc plus the capacity-first publish order in buffer_alloc() so the attribute stays sound. Spreading sized_by onto the pool/p_malloc helpers here would widen the surface (and the review) without being required for this buffer.

Happy to open a follow-up that annotates the malloc variants once this lands, if that direction still looks useful to you.

[LaptopsPlural]

Thanks again for the malloc-variants question.

I looked at folding that into this tip. The allocator APIs (i_malloc / p_malloc / t_*) are not a sized_by fit — that attribute is for ptr+capacity struct members, and those entry points would be a different (alloc_size) surface across several headers. The nearest internal companions (e.g. allocfree pool_block.block) are interior pointers into the same allocation, which I would rather not annotate here.

Prefer keeping this PR buffer-only (w_buffer↔alloc, capacity-first publish). Please confirm whether you want a specific malloc-side site folded into this PR, or are fine with a follow-up once this lands. Re-requesting review on the current tip either way.