Use `MLD_ALLOC` for all large allocations

[mkannwischer]

#801 added MLD_ALLOC for all large structs in sign.c. However, there are some larger buffers further down the call stack that we should consider also allocating via MLD_ALLOC. Ideally, allocations should be done on the higher level and work buffers be passed to the lower levels - but this will complicate CBMC proofs.

The following functions with large stack footprint would be a good place to start:

• mld_poly_uniform_4x: 3,456 bytes -

  mldsa-native/mldsa/src/poly.c

  Lines 665 to 666 in 4ab0e26

  	MLD_ALIGN uint8_t
  	buf[4][MLD_ALIGN_UP(MLD_POLY_UNIFORM_NBLOCKS * MLD_STREAM128_BLOCKBYTES)];

• mld_poly_uniform_gamma1_4x: 2,816 bytes -

  mldsa-native/mldsa/src/poly_kl.c

  Lines 531 to 532 in 4ab0e26

  	MLD_ALIGN uint8_t buf[4][MLD_ALIGN_UP(MLD_POLY_UNIFORM_GAMMA1_NBLOCKS *
  	MLD_STREAM256_BLOCKBYTES)];

• mld_poly_uniform_eta_4x: 1,152 bytes -

  mldsa-native/mldsa/src/poly_kl.c

  Lines 352 to 353 in 4ab0e26

  	MLD_ALIGN uint8_t buf[4][MLD_ALIGN_UP(MLD_POLY_UNIFORM_ETA_NBLOCKS *
  	MLD_STREAM256_BLOCKBYTES)];

• mld_poly_uniform: 840 bytes -

  mldsa-native/mldsa/src/poly.c

  Line 631 in 4ab0e26

  MLD_ALIGN uint8_t buf[MLD_POLY_UNIFORM_NBLOCKS * MLD_STREAM128_BLOCKBYTES];

• mld_poly_uniform_gamma1: 680 bytes -

  mldsa-native/mldsa/src/poly_kl.c

  Line 497 in 4ab0e26

  buf[MLD_POLY_UNIFORM_GAMMA1_NBLOCKS * MLD_STREAM256_BLOCKBYTES];