such separation serves multiple purposes:

- by having the common path for __tls_get_addr alone in its own
  function with a tail call to the slow case, code generation is
  greatly improved.

- by having __tls_get_addr in it own file, it can be replaced on a
  per-arch basis as needed, for optimization or ABI-specific purposes.

- by removing __tls_get_addr from __init_tls.c, a few bytes of code
  are shaved off of static binaries (which are unlikely to use this
  function unless the linker messed up).