Recover SV, ev method, backwards method

[felixhekhorn]

1. recover SV: a "major" change is to use always Lsv in the callbacks
2. recover evolution method (iterate-exact, etc.) - trivial thanks to Pass evolution + backwards method internally via int #395
3. recover backwards method - trivial thanks to Pass evolution + backwards method internally via int #395

[felixhekhorn]

cc @tgiani

the last few commits add the LHA benchmark using Rust and eventually I made it running: https://github.com/NNPDF/eko/actions/runs/10791934092/job/29930458728 - to be compared to the last run on master here https://github.com/NNPDF/eko/actions/runs/10683126160/job/29610483008

good things first:

• it is actually running ✔️ 🙈
• the benchmark now only takes 5 min as compared to 45 min ✔️ Both timings include compilation and I expect $t_{compile} < t_{run}$ for Rust, but inverse for Numba; recall that based on our previous observation we expect a run time reduction of ~4x
• the actual numbers are still mostly fine, BUT

the bad news are: they are not the same and even slightly worse ... compare

# small x gluon
0   1.000000e-07  10000.0  ...  1008.400000      -0.114663 # here
# vs.
0   1.000000e-07  10000.0  ...  1008.400000      -0.092259 # master

it's small, but still ... let's implement the missing unit tests in #385 and either we find something or I think we can accept ...

[tgiani]

@felixhekhorn indeed let me first complete the unit tests in #385, it is well likely that there is some little bug in as2.rs that will be spotted by doing that

[tgiani]

@felixhekhorn there was indeed a bug in as2. Still need to finish the unit test, maybe there will be more

[felixhekhorn]

@felixhekhorn there was indeed a bug in as2. Still need to finish the unit test, maybe there will be more

let's complete the unit tests, but the benchmark numbers seem now one on one to be the same: https://github.com/NNPDF/eko/actions/runs/10829978519/job/30048696220 🎉