OGL StreamBuffer optimization

[degasus]

OGL::StreamBuffer was very high in profilation (2.5%), so here are some optimization for it.

• Alloc fences directly in the StreamBuffer class, so they can't alias other member variables
• Align the size by power of two and use shifting instead of divisions for slot calculation
• Move alignment to the caller, so it can be skipped if not needed (eg indexbuffer)
• Reduce index size to 32bit, faster alignment division

[neobrain]

Don't use abbreviations in titles which show up in the commit log. Fix both the PR title and the commit messages.

Otherwise, LGTM.

[degasus]

@neobrain done