all: Make file-scope variables internally linked where applicable (#57) ...

Narrows the scope of visible symbols to where they're actually used. Also
makes it easier to see true globals in source files
(ones used from multiple translation units)

lock: Use stdatomic.h (#56) ...

Provides the same assembly output while using the standardized interface

e.g.

0000000000000000 <set_priv_smc_in_progress>:
   0:   90000000    adrp    x0, 0 <set_priv_smc_in_progress>
   4:   91000000    add x0, x0, #0x0
   8:   52800022    mov w2, #0x1                    // #1
   c:   d503201f    nop
  10:   085ffc01    ldaxrb  w1, [x0]
  14:   08037c02    stxrb   w3, w2, [x0]
  18:   35ffffc3    cbnz    w3, 10 <set_priv_smc_in_progress+0x10>
  1c:   72001c3f    tst w1, #0xff
  20:   54ffff81    b.ne    10 <set_priv_smc_in_progress+0x10>  // b.any
  24:   d65f03c0    ret

fuse: simplify checks in fuse_get_reserved_odm and fuse_get_spare_bit (#54) ...

A separate variable isn't particularly necessary here, and given that the type
is unsigned, we only need to test the upper bound.

This also resolves two -Wtype-limits warnings

mc: Remove dead code in get_carveout_by_id (#52)

se: Remove memset in se_perform_aes_block_operation (#53) ...

We can just initialize the array to be zeroed out. This is safer and less error-prone,
since the initializer is now associated with the variable directly, making it
impossible to put code relying on the zeroed out state before it (unlike with memset).

Failed attempt to fix loops not being unrolled, other fixes.

Move source files to src/, add Makefile, fix all build and linkage errors, etc.