Introduce "alloc" feature

[leonbotros]

Currently, this library uses the alloc feature of BLS12-381 (which is enabled by default) for some optimizations. I think we should introduce this as a default, but we should always have an option to not use allocation. This is useful when you compile this lib for an embedded target (i.e., a hardware module performing extraction).

Currently, the following allocating optimizations are used:

• some decryption algorithms perform the multi_miller_loop to compute multiple pairings.
• some setup/keygen algorithms use w-NAF precomputations.

[dobby-coder]

This issue has had no activity since 2021-11-01. Looking at the current repo, I don't see an explicit alloc feature flag in Cargo.toml, so the embedded-friendly split requested here doesn't appear to have landed.

If an embedded / no-alloc build is no longer a target for this crate, this can be closed. Otherwise, a short note confirming it's still wanted would help.

Dobby backlog maintenance 2026-04-16