feat: Add Narwhals/Ibis support for BE compatibility

[luizhsuperti]

Big project:
Currently cluster-experiments only accepts pandas DataFrames. tea-tasting has capabilities of
analyzing A/B analysis directly on data stacks (Polars, DuckDB, BigQuery, Snowflake) without pulling
data into pandas first.

Proposal

Integrate Narwhals to accept any
DataFrame-like object.

A separate, follow-up could add
Ibis support for true in-database
computation (BigQuery, Snowflake, ClickHouse, etc.).

Suggested scope (Phase 1 - Narwhals)

• Wrap public-facing analyze() and power_analysis() entrypoints with
  nw.from_native()
• Convert to pandas internally for now (zero-change to existing logic)
• Add Polars to the test matrix

Suggested scope (Phase 2 - Ibis)

• Push aggregations down to the backend before pulling into Python
• Only relevant for AnalysisPlan.analyze(), not simulation-based power analysis