feat(sigma): add Sigma Computing context-source adapter (#316)

* feat(sigma): add Sigma Computing context-source adapter

Closes #168

Adds a full ingest adapter for Sigma Computing so `ktx ingest` can pull
data model specs and workbook summaries into the ktx context layer. The
implementation follows the same fetch → chunk → project → LLM pattern
used by the Looker, Metabase, and MetricFlow adapters.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(sigma): address PR review comments

- Remove manifest from rawFiles; moves to peerFileIndex so fetchedAt
  changes don't mark all work units dirty every run
- Fix workbookFilter.updatedSince eviction bug: fetch full universe first,
  apply filter client-side, evict only on archived/deleted
- Remove measure projection entirely; project() writes measures: [] and
  the sigma_ingest skill surfaces Lookup/aggregation formulas as wiki prose
- Remove joins projection (v1 limitation); project() writes joins: [] and
  Lookup relationships are described in wiki prose instead
- Remove write-back dead code: createDataModel, updateDataModel,
  SigmaDataModelPushResult, mutate/post/put
- Fix emitBatches notes pluralization bug ('2 data modelss' → '2 data models')
- Add tokenInflight dedup on ensureToken to coalesce concurrent auth requests
- Retry spec fetch when existing staged spec is null (transient failure cache)
- Drop unused WorkbookFilter import from client-port.ts
- Note in docs that joins are not projected from Sigma data models in this release

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* updates

* fix(sigma): restore sigma in local adapter test + small cleanups

The gdrive↔sigma merge dropped 'sigma' from the expected adapter source
list in local-adapters.test.ts while keeping gdrive, so the slow TS suite
failed even though the source registers both. Add 'sigma' back at its
registration position (after metabase, before gdrive).

Also:
- Move the orphaned SigmaPullConfig docstring onto the schema it documents
  and drop the stale BullMQ reference (standalone ktx has no BullMQ; the
  config lives in the ingest job's bundleRef.config).
- Drop an O(n^2) find() round-trip in fetch() when building the active
  data-model list; filter once and reuse for the eviction id set.

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Andrey Avtomonov <andreybavt@gmail.com>
Co-authored-by: Luca Martial <48870843+luca-martial@users.noreply.github.com>

packages/cli/test/context/memory/memory-runtime-assets.test.ts

@@ -24,6 +24,7 @@ const expectedAdapterSkillHeadings: Record<string, string> = {
   lookml_ingest: '# LookML to ktx Semantic Layer',
   metabase_ingest: '# Metabase to ktx Semantic Layer',
   metricflow_ingest: '# MetricFlow to ktx Semantic Layer',
+  sigma_ingest: '# Sigma Ingest',
 };
 const verificationWriterSkills = [
   'gdrive_synthesize',
@@ -33,6 +34,7 @@ const verificationWriterSkills = [
   'looker_ingest',
   'metabase_ingest',
   'metricflow_ingest',
+  'sigma_ingest',
   'live_database_ingest',
   'historic_sql_table_digest',
   'historic_sql_patterns',