Add wandb logging to BootstrapFewShot's metric_val

[CShorten]

Weights & Biases (wandb) are one of the industry leaders in experiment tracking software for Machine Learning.

I think runs and sweeps both offer significant value to monitoring DSPy optimization runs.

This PR will add wandb logging to BootstrapFewShot to log the values of metric_val as the teacher model is bootstrapping traces through the program.

Intended use:

# connect to wandb project
wandb.init(
  project: "my-dspy-program",
  config = {
    "program": "Perspective-Guided-STORM",
    "teacherLM": "GPT4"
  }
)

# wandb logging is internal to the `teleprompter.compile` call
# ^ you just add values to the project with wandb.log(...)

compiled_progarm = teleprompter.compile(PG_STORM, trainset=trainset, ...)

I think this will be especially effective when say looping through program configurations (in terms of multi-model design or the composition of modules) or teacherLMs.

As a result, DSPy users will have a nice visualization of how effectively the teacher LM is able to bootstrap traces through the program that surpass the metric_threshold.

I think this is a nice building block, but generally the wandb integration will be more effective with higher-level optimizers such as BootstrapFewShotRandomSearch, BootstrapFewShotWithOptuna, or the BayesianSignatureOptimizer.

[isaacbmiller]

Maybe better to implement this in config so we don't have to pass it as an individual parameter to each class and child?

[CShorten]

Just in case anyone had seen this --> moved here: #849