Add SDK support for bounded Snowflake destination queries

[pkozikow]

Track Python SDK propagation of POST /v1/destinations/{destinationId}/query from Platform #19472, paired with Sonar #6544. This operation is separate from workspace skill documentation tracked in #211.

Regenerate from the updated public API definition and expose the operation in Destinations (publicQueryDestination, SDK alias queryDestination). Requests contain exactly one statement string, limited to 32 KiB UTF-8 within a 64 KiB JSON body.

The response contains ordered columns (index, name, type, nullable), positional rows of string-or-null cells, and a page with returnedRows, fixed rowLimit: 1000, fixed byteLimit: 1048576, truncated, an array of row_limit/byte_limit/provider_partition reasons, and required nextCursor: null. Preserve empty arrays, duplicate column labels, null cells and provider text without automatic native-value coercion. The full response is at most 1 MiB with at most 256 columns. There is no resumable cursor or total-row count.

Acceptance:

• Generate the operation and request/response models, and add serialization tests for positional null cells, duplicate labels, empty rows/reasons and the required null cursor.
• Verify all documented 401/403/404/413/422/501/502/503/504 problem responses.
• Ensure this query operation does not automatically retry an uncertain POST submission, including timeout, disconnect or retryable HTTP status; preserve other operations' existing retry behavior.
• Document workspace reader access, controlled availability, key-pair Snowflake destinations and separately approved query-role configuration. Other warehouse/auth variants are unsupported; availability is disabled by default.
• Release SDK support after the paired server change is ready, then link the released version here and in the Platform PR. SDK release and production activation are separate.