feat(dataframe): expose introspection methods (schema, explain, cache, describe)

[andygrove]

Is your feature request related to a problem or challenge?

DataFrame introspection is currently limited to count(), show(),
and show(int). Users wanting to inspect schema, see the planned
query, or materialize an intermediate result have no Java entry point.

Describe the solution you'd like

• DataFrame.schema() — return an Arrow Schema. Reuse the IPC
  round-trip already established for SessionContext.tableSchema
  (tableSchemaIpc pattern in SessionContext.java). Non-consuming;
  the DataFrame remains usable.
• DataFrame.explain(boolean verbose, boolean analyze) — wraps
  DataFusion::DataFrame::explain, returning a DataFrame whose rows
  are the plan-explanation strings. Caller calls show() / collect()
  on the result. Matches DataFusion's own semantics.
• DataFrame.cache() — materializes the plan into an in-memory
  table and returns a new DataFrame. Async on the Rust side; blocks on
  the Tokio runtime, same pattern as collect. Caller-closes the
  returned DataFrame.
• DataFrame.describe() — async, returns a DataFrame with summary
  stats (count, mean, stddev, min, max) per numeric column. Same
  pattern as cache.

Describe alternatives you've considered

For schema: ctx.tableSchema(name) works only for registered
tables. A user who built a DataFrame via sql("SELECT …") or chained
transformations has no schema accessor.

For explain: ctx.sql("EXPLAIN <query>") works but only against a
SQL string.

Additional context

All four are non-consuming except cache and describe (which return
new DataFrames the caller owns and closes). The schema/explain pair are
the most-requested and could land first as a smaller PR; cache and
describe are independent and can follow.