feat: add namespace insert write path

[jackye1995]

Adds an opt-in namespace insert append path for namespace-backed Lance tables.

Expected user experience:

• Existing INSERT INTO and append writes keep using the default Spark writer unless users set use_namespace_insert=true.
• With namespace insert enabled, users still submit a normal DataFrame append; Spark runs executor-side writer tasks that send Arrow batches through the Lance namespace insert API.
• namespace_insert_parallelism lets users request the number of writer tasks; sharded tables use the sharding distribution, and unsharded tables repartition by the first output column.
• Directory and REST namespaces use the same API path, so REST namespaces can handle ingestion without changing Spark user code.
• Insert requests commit as they run, so users should use the default writer when they need Spark driver-side atomic commit behavior.

Includes local/rest-dir Docker CI coverage and documentation updates.