fix(bind): accept dictionary-encoded bind parameter columns (CONV-2)

[fornwall]

Fixes the gap covered by fornwall/validation#6 (TestStatement.test_parameter_dictionary_encoded): binding a dictionary-encoded column — what pandas categoricals produce over the C data interface — failed InvalidArguments with unsupported Arrow type Dictionary(Int32, Utf8) even though the value type is supported (REVIEW.md CONV-2).

What changed

• cell_value gets a Dictionary arm (src/bind.rs): dictionary encoding is an index representation of the same logical values, not a different logical type (the Arrow columnar format's "Dictionary-encoded Layout"), so the column binds transparently as its value type. The key at each row selects the dictionary value, which re-enters the same cell mapping — every bindable value type is accepted encoded, scalars and ARRAY<...> alike, and an unsupported value type is rejected with the same error as its plain form (on null cells too, via null_dictionary_value, keeping the fail-loudly-on-every-row convention the Decimal128 scale check set).
• Mutation-based bulk ingest decodes the same way for free — insert_mutation shares cell_value.
• spanner_column_type sees through the encoding, so ingest create modes map a dictionary column to its value type's Spanner column type.
• RunEndEncoded (the other encoding CONV-2 mentions) stays rejected — no known producer emits it over the C data interface today; noted in REVIEW.md.

Verification

• The origin validation test, run against the emulator via scripts/run-foundry-validation.sh pointed at test(statement): cover dictionary-encoded bind parameter columns validation#6's head: fails with the exact CONV-2 error before this change, passes after (verified both directions).
• Full scripts/with-emulator.sh cargo test green (288 + 52 + 17 + 6 + 1).
• Full foundry suite at its pinned upstream ref: 184 passed, 65 skipped.
• cargo clippy --all-targets --all-features -- -D warnings and cargo fmt --all --check clean.
• New unit tests cover decoded scalar + widened-int binds, a List-valued dictionary, null cells, sliced batches (non-zero key offsets), unsupported value types (null cells included), ingest mutations, and the DDL mapping.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FjqdweBVqxdegAHVDh97oV