Add parameter-aware bulk inserts

[elpete]

Addresses #153

What changed

• add an insertBulk() API for arrays of structs
• use one serialized JSON parameter with OPENJSON for SQL Server bulk inserts
• infer SQL Server column types using the same type-resolution foundation as whereInBulk, with optional sqlTypes overrides
• preserve SQL Server returning() support through the OUTPUT clause
• use parameter-aware multi-row insert batches as the portable fallback for grammars without a native strategy
• accept a positive chunkSize as an explicit cap; consistent with pagination conventions, non-positive values mean all rows subject to fallback grammar parameter limits

Why

Large multi-row inserts can exceed database parameter limits. SQL Server can avoid that limit entirely by sending the rows as one JSON binding and expanding them into a typed rowset with OPENJSON. Other grammars continue to receive safe batching without exposing chunking as the primary API.

Tests

Tests were added first for explicit fallback batching, grammar-limit enforcement, empty input, the non-positive all-rows convention, SQL Server OPENJSON, inferred and explicit SQL types, serialized bindings, and returning().

• TestBox on Lucee 6: 2,779 passed, 0 failed, 3 skipped
• box run-script format:check
• git diff --check

[elpete]

Refactored the bulk insert boundary after review: QueryBuilder now has a preparation method that returns columns and the JSON binding, while SQL generation remains exclusively in the grammar’s compileBulkInsert() method. Generated SQL and binding behavior are unchanged.

Validation: Lucee suite 2,781 passed, 0 failed/errors, 3 skipped; formatting and git diff --check pass.