refactor: migrate all notebooks from pandas to polars

[raulk]

Summary

Migrates all data transformation logic from pandas to polars for improved performance and more expressive data manipulation.

Base branch: feat/network-overview (PR #43)

Changes

• Add polars>=1.0 to dependencies
• Migrate all 9 notebooks to use polars for data transformations
• Keep pandas only where needed for plotly compatibility (.to_pandas())

Migration pattern

# Before (pandas)
df = load_parquet("dataset", target_date)
df_grouped = df.groupby("col").agg({"value": "sum"})

# After (polars)
df = pl.from_pandas(load_parquet("dataset", target_date))
df_grouped = df.group_by("col").agg(pl.col("value").sum())
fig = px.bar(df_grouped.to_pandas(), ...)  # Convert for plotly

Key polars patterns used

pandas	polars
df.groupby().agg()	df.group_by().agg()
df[df["col"] > 0]	df.filter(pl.col("col") > 0)
df["col"] = ...	df.with_columns(...)
df.sort_values()	df.sort()
df.drop_duplicates()	df.unique()
df.fillna()	df.fill_null()
df["col"].map({...})	pl.when().then().otherwise()
df.merge()	df.join()
df.pivot()	df.pivot()
df.melt()	df.unpivot()

Benefits

• Performance: Polars is significantly faster than pandas for large datasets
• Memory efficiency: Lazy evaluation and better memory management
• Expressiveness: Chainable API with clear intent
• Type safety: Better handling of nulls and type conversions

Test plan

• Run just fetch <date> to fetch data
• Run just render <date> to verify all notebooks render correctly
• Verify visualizations display correctly in rendered HTML