reader(csharp): SEA cannot force inline results — cloudfetch.enabled=false is silently ignored #571

Problem (known limitation)

On the SEA/REST path, adbc.databricks.cloudfetch.enabled=false is silently ignored — there is currently no reliable way to force pure inline (non-CloudFetch) results, unlike Thrift.

Evidence

• Thrift disables CloudFetch properly: cloudfetch.enabled=false → DatabricksStatement.cs:478 statement.CanDownloadResult = useCloudFetch → server returns inline results.
• SEA has no CanDownloadResult equivalent. UseCloudFetch is explicitly accept-but-ignore in StatementExecutionStatement.SetOption (csharp/src/StatementExecution/StatementExecutionStatement.cs:252-257) and is never read by StatementExecutionConnection. Inline-vs-external is governed solely by adbc.databricks.rest.result_disposition (default INLINE_OR_EXTERNAL_LINKS, StatementExecutionConnection.cs:331).
• The disposition=INLINE workaround is not a real substitute: INLINE is server-capped at 25 MiB (DatabricksParameters.cs:419), and the inline JSON data_array path is unimplemented (the DataArray model fields have zero readers; design-doc TODO at csharp/doc/statement-execution-api-design.md:2657-2658). Inline Arrow is implemented and multi-chunk (InlineArrowStreamReader, FetchAllChunkSegmentsAsync, StatementExecutionStatement.cs:843-962).

Impact

Matters for network-restricted deployments (egress to S3/Azure Blob blocked, restrictive proxy) where a client must force inline to avoid presigned cloud-storage URLs. Reyden is unaffected (never emits external links → always inline). This only bites standard SQL warehouses over SEA.

Fix dependency / proposed action

Full inline forcing on SEA depends on the server returning Arrow in Attachment for INLINE disposition (the "inline-Arrow" server capability). The client-side reader (including multi-chunk) already exists. Once inline-Arrow lands, map cloudfetch.enabled=false → disposition=INLINE on the SEA path to restore Thrift parity. Until then, document this as a known SEA limitation.