use gollm=>mira equation to amr pipeline

[mwdchang]

Summary

Replace latex => sympy.parse_latex => mira pipeline with latex => gollm => mira pipeline.
This is the 2nd part to: #5846

This PR also clean up the mira-endpoint.
/mira/latex-to-amr => /knowledge/equations-to-model-debug

Testing

POST /knowledge/equations-to-model should return a model-id, which in turn should reflect the equations.

{
  "equations": [
  "\\frac{d S(t)}{d t} = - b_q S(t) I(t)",
  "\\frac{d E(t)}{d t} = b_q S(t) I(t) - r_X E(t)",
  "\\frac{d I(t)}{d t} = r_X E(t) - g_p_q I(t)",
  "\\frac{d R(t)}{d t} = g_p_q I(t)"
   ]
}

[YohannParis]

Lovely!

[Tom-Szendrey]

why the debug in equations-to-model-debug ?

[mwdchang]

why the debug in equations-to-model-debug ?

@Tom-Szendrey I want to flag it as not meant to be used aside from debugging. The current pipeline is not very transparent, it has 3-4 junctions (llm-enrichment, llm-code-gen, code-eval to sympy, ode-to-amr) where things can go wrong but not really set up to report errors at each stage. Because it is also coupled with SKEMA stuff and model enrichment, and is used as a one-shot model creation process in the UI, it is just not ideal to refactor it until we agree on a more sane process. So for now we have a debugging analog.

[Tom-Szendrey]

why the debug in equations-to-model-debug ?

@Tom-Szendrey I want to flag it as not meant to be used aside from debugging. The current pipeline is not very transparent, it has 3-4 junctions (llm-enrichment, llm-code-gen, code-eval to sympy, ode-to-amr) where things can go wrong but not really set up to report errors at each stage. Because it is also coupled with SKEMA stuff and model enrichment, and is used as a one-shot model creation process in the UI, it is just not ideal to refactor it until we agree on a more sane process. So for now we have a debugging analog.

Thank you for filling me in!