example of starting ax experiment from dataframe data (#5098)

Summary:
Pull Request resolved: #5098

This diff adds a new Ax tutorial notebook demonstrating how to initialize an experiment from pre-existing DataFrame data.

**Key Features Covered:**
1. **Attaching Historical Trials** — Shows how to use `client.attach_trial()` and `client.complete_trial()` to import historical experiment data (arms with parameters and their observed metric values) from a pandas DataFrame
2. **Warm-Starting Bayesian Optimization** — After attaching historical data, uses Ax's `Client` to generate new candidate trials with model-based optimization (BoTorch)
3. **Analysis & Visualization** — Demonstrates Ax's built-in analysis tools: cross-validation plots, utility progression tracking, and arm effects visualization (both observed and predicted)

**Use Cases:**
- Migrating experiments run outside of Ax into the Ax framework
- Leveraging existing CSV/database data to warm-start Bayesian optimization
- Building on historical configurations and outcomes

**Technical Details:**
- Uses the Branin function as a benchmark optimization problem (2 parameters, known global minimum ≈ 0.398)
- Creates 15 quasi-random historical evaluations and attaches them as completed trials
- Generates and evaluates a new candidate using `client.get_next_trial()`
- Compares predicted vs. observed arm effects

Reviewed By: andycylmeta

Differential Revision: D93760064

fbshipit-source-id: 7516ebc70ae666d7e264c8ab7e249fff38f7a244

Author: sdaulton