Review whether lazy binding is necessary in rtdl

[64]

We should review whether lazy binding (RTLD_LAZY) is necessary to achieve good performance on large binaries. If not, we should remove it from the code.

If we assume that performance on-par with glibc is achievable, it's probably enough to run some programs with different values of LD_BIND_NOW and see how much that affects things. Alternatively, we could do some profiling with mlibc-lfs.

[qookei]

Most Linux distributions these days link all their packages with -z now, so I think it's safe to say that lazy binding support can be removed. Our current code for it is kinda broken anyway (not implemented at all on AArch64, and on x86_64 we don't stash SSE registers as we ought to).