Add generic query-catalog and source-worker publishing flow

[lupuszr]

Summary

• add generic QueryCatalog support to v9 profile loading and UI rendering so saved queries can be grouped and displayed as first-class profile resources
• introduce reusable transducer, share-batching, source-registry, and source-worker plumbing, plus dkg source-worker run --config ... [--once], to support generic async source ingestion workflows
• harden async publishing with stale-worker fencing coverage and related targeted tests/build validation

CLI Usage
The new source-worker entrypoint runs a generic ingestion handler from a config file and pushes work through the existing async publish flow.
Run continuously:

dkg source-worker run --config /path/to/worker.config.json

Run once for cron/manual execution:

dkg source-worker run --config /path/to/worker.config.json --once

The worker config points the CLI at:

• a handler module and export to load source-specific ingestion logic
• a daemon URL/token for SWM writes and async job polling
• a state file for persistence across runs
  This keeps the worker runtime generic in dkg, while allowing source-specific adapters to live outside the core repo.

{
  "daemonUrl": "http://127.0.0.1:9200",
  "daemonToken": "your-daemon-token",
  "stateFile": "./.tmp/source-worker-state.json",
  "pollIntervalMs": 30000,
  "handlerModule": "/absolute/path/to/source-worker-handler.js",
  "handlerExport": "createSourceHandler",
  "sources": [
    {
      "id": "example-source",
      "kind": "custom",
      "enabled": true,
      "config": {
        "inputPath": "/data/example.csv",
        "dataset": "example-dataset"
      }
    }
  ]
}

[github-actions]

🔴 Bug: This only fences the update() path. A worker that loses its claim while publishExecutor is running can still mutate the job through recordPublishResult() / recordPublishFailure(), because those methods still write state without re-checking the wallet lock / claim token. Please apply the same fence validation before every post-claim mutation and add regression coverage for the publish-result/failure paths.

[github-actions]

🔴 Bug: no-matching-rows is treated as a success state, but this loop only derives success from lastJobIds. If a handler returns { lastStatus: 'no-matching-rows' } with no jobs, the next poll sees aggregate === '', falls through here, and keeps retrying until the source is incorrectly forced into manual review. Short-circuit same-fingerprint no-matching-rows before incrementing retries (or persist a synthetic terminal status into the aggregate calculation).

[github-actions]

🔴 Bug: This reads only config.chain?.tokenAddress, so a token override coming from the merged network config is ignored here. In that case /api/wallets/balances still falls back to Hub.Token, which can make the UI show the wrong token or fail entirely while dkg wallet succeeds. Use the resolved chain/network token address in this route too.

[github-actions]

🔴 Bug: processSource() gets the computed fingerprint, but the worker persists result.nextState verbatim. If a handler omits that field from nextState (which is easy with the current API), the next poll sees current.fingerprint as missing and reprocesses the same source again, potentially creating duplicate share/publish jobs forever. Merge the framework-owned fields here (fingerprint, lastRunAt, and any retry/manual-review resets) instead of requiring every handler to remember them.

[github-actions]

🔴 Bug: rank comes from Date.now() above, but this writes it as xsd:int. Millisecond timestamps are already around 1.7e12, far beyond the 32-bit range of xsd:int, so the stored literal is invalid/overflow-prone and query ordering can become store-dependent. Persist this as xsd:integer/xsd:long, or use a smaller bounded rank value.

[github-actions]

🟡 Issue: This concatenates localSavedCatalogs and profile.queryCatalogs without merging by catalog identity. If the persisted profile already contains ui-saved-queries and the user saves another query, the UI renders duplicate catalog sections with the same React key and splits the queries across them. Merge catalogs by subGraph|slug before rendering.