secp256k1 precompiles: add, double and decompress + refactor

[mcalancea]

Note: this PR is done against Inversed's fork master, do not merge.

• adds secp256k1_add, secp256k1_double and secp256k1_decompress precompiles with the same interface as sp1, but backed by the secp crate.
• integrates appropriate dummy circuits for precompiles into the zkvm
• added utils.rs with some utilities for manipulating memory segments of the VM.
• tested the compatibility of the syscalls (+ keccak, done previously) with sp1 (see the syscalls.rs example file).
• extended the docs pasted from sp1

Questions:

• sp1 does one additional check for syscalls that only take one argument, namely that the second argument/register is 0. I haven't done this yet; do we want to? Or can we be more permissive?
• sp1 does generally require alignment of the pointers along a 4-byte boundary. I think it may be helpful to require stricter alignment for arrays of larger types (for example for keccak which inputs u64s, it could be aligned to 8 bytes).

[naure]

the second argument/register is 0

For the purpose of compatibility and avoiding surprises I think we should do that too, with a comment to remember that it is the only reason.

alignment of the pointers along a 4-byte boundary

Ceno memory system works with 4-bytes words on 4-bytes alignment; there are no benefits with 8.

[naure]

Great work with batteries included!

• Witgen & Runtime integration
• Refactoring & Clarity
• Specs & Refs & Comments
• Tests & Examples & Compatibility

➜ I think there are some memory ops missing, see comments.