Make p_vaddr % p_align == p_offset % p_align for (some) TLS segments.

Make p_vaddr % p_align == p_offset % p_align for (some) TLS segments.

See https://sourceware.org/bugzilla/show_bug.cgi?id=24606 for the test case.
See https://reviews.llvm.org/D64930 for the background and more discussion.

Also this fixes another bug in malloc_aligned() where total size of
the allocated memory might be not enough to fit the aligned requested
block after the initial pointer is incremented by the pointer size.

Reviewed by:	bdragon
Tested by:	antoine (exp-run PR 244866), bdragon, emaste
Sponsored by:	The FreeBSD Foundation
MFC after:	2 weeks
Differential revision:	https://reviews.freebsd.org/D21163