docs(datatypes): Parquet BYTE_ARRAY strings and binaries read as LargeUtf8/LargeBinary (#2252) Co-authored-by: claudespice <claudespice@users.noreply.github.com>

docs(reference): document the sub-second duration units and fractional values The Duration reference listed only s/m/h/d/w and described durations as integers. Both parsers the runtime uses accept more: duration-parse (used by spicepod runtime fields) and fundu 2.0.1 (used by acceleration duration params) both accept ns, microseconds, and ms, accept fractional and exponent forms, treat a bare number as seconds, and reject month/year. fundu is pinned at 2.0.1 in every release tag from v1.5.2 through v2.1.1 and on trunk, so this correction applies to all versioned snapshots.

docs(datasets): a time_format mismatch fails the dataset, and iso8601 on a timestamp or date column is ignored with a warning (#2325) * docs(datasets): a time_format mismatch fails the dataset, and iso8601 on a timestamp or date column is ignored with a warning (spiceai/spiceai#14777) * docs(datasets): a string time_column needs time_format iso8601 (spiceai/spiceai#14777)

docs(sql): SQL PREPARE/EXECUTE need a Flight SQL session; document parameter binding (#2251)

Add versioned docs for v2.1 (#1903) Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

fix: add the missing Elasticsearch row to the Distribution Availability table (fixes #2208) (#2228) * docs(distributions): add the missing Elasticsearch row to Distribution Availability The page enumerates the connectors gated out of the default `spiced` build twice. "Additional Connectors" names four (ODBC, ScyllaDB, NFS, Elasticsearch); "Distribution Availability", the table a reader scans first, carried rows for only three of them. A reader checking whether the stock open source image can talk to Elasticsearch found the connector absent from the availability table and could reasonably conclude it ships by default. It does not. * docs(distributions): carry the Elasticsearch row into the v2.0.x-v2.2.x snapshots The connector exists and is gated out of the default `spiced` build in every release from v2.0 onward, so the same omission is present in those snapshots' availability tables, and their Additional Connectors section named ODBC alone. v1.11.x is unchanged: that release has no Elasticsearch connector to document.

docs(file formats): json_pointer routes a jsonl dataset through Spice's JSON reader (#2140) * docs(file formats): json_pointer routes a jsonl dataset through Spice's JSON reader * docs(file formats): cite the version-specific source for the jsonl json_pointer routing Each versioned page now links the format-dispatch code for its own release line. The path is not the same across them — v2.0 and v2.1 carry it at crates/runtime/src/dataconnector/listing/connector.rs, and v2.2 onward at crates/data-connector-api/src/listing/connector.rs — so a single link would have been wrong on three of the four pages. --------- Co-authored-by: claudespice <270518434+claudespice@users.noreply.github.com>

Add versioned docs for v2.1 (#1903) Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Add versioned docs for v2.1 (#1903) Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Add versioned docs for v2.1 (#1903) Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

docs(performance-tuning): correct Cayenne footer-cache default (unset -> 50 MB, not 128 MB) The Performance Tuning guide stated the Cayenne footer cache defaults to 128 MB. Since v2.0, cayenne_footer_cache_mb is Option<usize> = None, and when unset the runtime applies DataFusion's 50 MB file-metadata-cache limit (footer_cache_mb: Option<usize> in crates/cayenne). This matches the Cayenne accelerator page, which was already corrected. Fixed in vNext + 2.1.x + 2.0.x; 1.11.x keeps 128 MB (correct for that release, before the field became Option).

Add versioned docs for v2.1 (#1903) Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

docs(task_history): the refresh task was renamed acceleration_refresh in v2.0 The task_history reference, the troubleshooting guide and the Zipkin page show sample runtime.task_history output in which the accelerated-table refresh rows carry the task name accelerated_refresh. That name was correct up to v1.11: the span was renamed to acceleration_refresh at v2.0.0, so every 2.x snapshot and vNext have been showing a task name no 2.x runtime emits. Scoped to website/docs and versioned_docs/version-2.*; the 1.5.x-1.11.x snapshots keep accelerated_refresh, which is what those releases record. cli/reference/trace.md already made this same split.