fix: improve Docker image publish speed

[daniilrrr]

Ticket

• Related Linear Ticket: SEQ-1049

What does this PR do?

• Summary:
  • CI go brr
• Key Changes:
  • Correct docker buildx... command for our multi-target Dockerfile - no more for loop. 6x faster :)
  • Removed unused steps from the Dockerfile, including SP1 toolchain.
  • Small code change to /shared so that no part of the /synd-withdrawals directory nor the heavy ZK deps get pulled into other services. ZK deps like sp1 need to stay isolated to synd-withdrawals
  • test-framework module refactor to build fewer binaries. See below
• <Goal of this section: someone reads this in a few months and at-a-glance understands why this change got made + what it is>

Breaking changes?

• No

Metrics changes?

• No

Anything interesting?

I discovered that cargo has special behavior for test files in a /tests directory. It will compile each test file into a separate binary. With many test files, that compilation overhead + all those artifacts add up. We can get around this via one main.rs test file which produces one binary for a faster+smaller build.
References:

• Blog from former cargo maintainer
• Specific PR where they did this within cargo
• Sample PR from a third party project making this same change

How can this PR be tested?

• CI

[linear]

SEQ-1049 Reduce build times for Docker images

use rust-cache (or similar) in our github actions to speed up our build times

also explore decoupling the withdrawals part

[daniilrrr]

building just translator took me 12min locally with Orbstack.

docker build . --no-cache --target synd-translator

[+] Building 720.0s (14/14) FINISHED                                                                                                                      docker:orbstack

[daniilrrr]

[jorgemmsilva]

this is crazy! great improvement!
⚡ ⚡ ⚡