feat(datasets)!: remove datasets commands and dataset feature surface

[eddietejeda]

Summary

Removes the datasets feature from the CLI entirely (full feature removal across code + docs).

Net: 17 files changed, ~126 insertions, ~971 deletions. Clean build, clippy clean, 194 unit + 12 integration tests passing.

What changed

Code

• Deleted src/datasets.rs (441 lines) and the datasets command group (command.rs enum + main.rs handler + module decl/import).
• indexes — dropped the --dataset-id scope everywhere: removed IndexScope::Dataset, list_one_dataset, the dataset path arms, and its unit test; simplified the create/delete/list handlers to connection/catalog scope only.
• jobs — removed dataset_refresh and create_dataset_index from parse_job_type and the --job-type value list.
• sdk.rs — the datasets() client lives in the external hotdata crate (not editable here), so its one X-Workspace-Id-header regression test was repointed off datasets().list onto jobs().list (another generated-client workspace-scoped call) to preserve the assertion.
• databases.rs — reworded the parquet-only error hint that pointed at hotdata datasets create.
• tests/workspace_env.rs — repointed the workspace-lock test from datasets list to indexes list.

Docs

• README + all skill docs scrubbed of the datasets command, --dataset-id, the removed job types, and datasets.main.* query examples (replaced with <catalog>.public.*). "Datasets vs managed databases" sections collapsed to managed-databases-only with consistent anchors.
• Generic-English "dataset" usage in the geospatial skill was intentionally left untouched.

Notes for reviewers

• CHANGELOG.md not hand-edited — the changelog is generated from conventional-commit messages at release time; this PR's feat(datasets)!: commit carries the breaking-change note.
• Pre-existing doc bug also fixed: several index docs showed hotdata indexes create --connection-id ..., but create has no --connection-id flag (the binary rejects it). Index-on-a-connection-table is done via --catalog <connection-name-or-id> (resolve_connection_id resolves names/IDs/catalog aliases). Since the dataset removal forced rewriting the "index scopes" prose anyway, these were corrected to --catalog rather than shipping broken examples.

Test plan

• cargo build — clean, no new warnings
• cargo clippy --all-targets — clean (only pre-existing style nits in untouched code)
• cargo test — 194 unit + 12 integration tests pass
• Smoke test: hotdata datasets rejected as unrecognized; indexes list has no --dataset-id; jobs --job-type shows only data_refresh_table, data_refresh_connection, create_index

🤖 Generated with Claude Code

[claude]

Clean feature removal. Code changes (command/main/indexes/jobs/sdk) are internally consistent — match arms stay exhaustive, error messages preserved, job display unaffected by the parse_job_type narrowing, and no dangling dataset references remain. Docs scrubbed thoroughly and the --catalog connection-scope correction matches the resolve_connection_id code path.

[codecov]

Codecov Report

❌ Patch coverage is 18.51852% with 22 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/main.rs	0.00%	22 Missing ⚠️

📢 Thoughts on this report? Let us know!

[claude]

Clean feature removal. Verified no leftover dataset references in src/ (module, command enum, handlers, IndexScope::Dataset, job types all removed), docs scrubbed consistently, and the repointed sdk.rs/workspace_env.rs tests preserve their original assertions. The jobs().list repoint matches the 4-arg signature, and indexes create/list retain prior connection-scope behavior.

[claude]

Clean, well-scoped removal of the datasets feature. Verified: no leftover dataset references in code/tests; indexes::list match remains exhaustive with the workspace-scan fallback intact; indexes create/delete collapse to connection scope with unchanged resolution behavior and the public schema default preserved; the sdk.rs workspace-header regression test was correctly repointed to another workspace-scoped call. Docs updated consistently.