ml-dsa: activate MaybeBox with alloc feature

[baloo]

When the matrices were pushed on the stack in #1344 and #1345, it relies on the MaybeBox implementation of module-lattice.

MaybeBox will push the matrix on the heap only if alloc is enabled.

Because alloc never got activated in ml-dsa this effectively reverted the optimization on the stack size made in #1261 and #1259.

This commit fixes that by enabling alloc on module-lattice when alloc is enabled on ml-dsa and restoring the expected behavior of MaybeBox.

[baloo]

a workaround for a consumer crate until this makes into a release is to enable the feature on module-lattice manually:

[dependencies]
ml-dsa = { version = "0.1", features = ["alloc"] }
module-lattice = { version = "0.2.3", features = ["alloc"] }

[tarcieri]

This looks like a dup of #1365? I think if you rebase on master it won't have any changes.

[baloo]

indeed, sorry for the noise.

[tarcieri]

I can cut a patch release with the fix

[baloo]

I think we can use the workaround, if you have time that'd be appreciated, but it's fine otherwise.