BufferVec: always allocate a GPU buffer

[SludgePhD]

Objective

When BufferVec::write_buffer is called without any data in the buffer, it does not allocate a GPU buffer, which then causes draw calls using that buffer to get skipped (usually). If the draw call does important work even with zero elements in the BufferVec, such as clearing the render target, this work is skipped as well, resulting in incorrect output.

Solution

It is unintuitive that BufferVec behaves like this, so this PR changes it to always allocate a GPU-side buffer when write_buffer or reserve is called, even when that buffer ends up with a size of 0.

---

Changelog

Fixed

Fixed BufferVec::write_buffer and BufferVec::reserve not allocating a GPU-side buffer when empty.

[JMS55]

What is your use case for an empty BufferVec? It looks like we currently only use it for skinning and morph target data, but that's going to go away soon anyways.

[SludgePhD]

I use it for an instance-rate vertex buffer to store instance data in. Should I be using something else for that? I briefly looked at DynamicStorageBuffer but it seems to have the same bug.

[jgayfer]

Hard to test with merge conflicts, but I've run into this same issue. It was a big gotcha for me that using a GpuArrayBuffer (which is backed by a BufferVec) would end up with no binding in the event of an empty array.

@SludgePhD I will gladly test + review once conflicts are fixed!

[SludgePhD]

Rebased it

[jgayfer]

I'm noticed that since this PR was opened, we now have RawBufferVec and BufferVec. In my case I was having issues with BufferVec as its used by GpuArrayBuffer.

Would it make sense to apply this change set to both variants?

[SludgePhD]

DynamicStorageBuffer has the same issue. If this is going to be fixed (I think it should), then it should probably be fixed for all of them. That said, I'm kinda out of spoons at the moment, so I'll close this PR. Feel free to take over!