implement the med xpert qa text scenario in the medhelm

[chakravarthik27]

This pull request introduces the MedXpertQA Text benchmark to the codebase, enabling the evaluation of medical question answering capabilities in large language models. It includes the implementation of the scenario, integration into run specifications, and updates to the configuration and dependencies to support the new benchmark. The most important changes are summarized below:

New Scenario Implementation:

• Added the MedXpertQATextScenario class in medxpert_qa_text_scenario.py, which loads and processes the MedXpertQA Text dataset from HuggingFace, structures instances for evaluation, and provides scenario metadata.

Integration with Benchmarking Framework:

• Registered a new run specification function get_medxpert_qa_text_spec() in medhelm_run_specs.py to define how the scenario should be run, including adapter and metric specs.

• Updated schema_medhelm.yaml to add medxpert_qa_text to the list of run groups and provided its display name, description, metric groups, environment, and taxonomy information for the benchmark schema.
  Dependency and Build Updates:

• Relaxed and aligned version constraints for several dependencies in pyproject.toml, such as datasets, numba, and together, and added tiktoken as a new dependency to support the new scenario.

• Pinned the setuptools version below 82 for openai-whisper extra build dependencies to avoid build issues.

[blidiselalin]

1. Re-pin numba and together in requirements.txt (don't leave them fully unpinned).
2. Clarify whether tiktoken should be core vs. optional.
3. Fix the citation and description mismatch in schema_medhelm.yaml

[blidiselalin]

Looks good