Deepen the scope of the data sources

[Paruccone]

it seems to me (can't verify because of issue [19]) that the data sources do not take into account queries that are not loaded (but make it to the final semantic model because they are merged/appended to exposed queries).

Would be useful to have an option to include them.

[JonathanJihwanKim]

Resolved in ed986ed. You were exactly right — the catalog was only inspecting partition M of loaded tables and missing connector calls buried inside shared expressions (Enable Load = false, helpers consumed via Table.Combine/Table.NestedJoin, etc.).

What changed:

• m-parser.js now performs a second pass that scans every shared M expression for connector calls, skipping parameters (IsParameterQuery=true), functions, lists, and records.
• Discovered sources are tagged isNonLoadedQuery=true with the originating expressionName.
• The lineage engine builds a transitive expression-reference graph so each non-loaded source can list the loaded tables that ultimately consume it (via merge/append/delegation).
• The Markdown and HTML exports show a "Non-loaded query: `name`" badge on these sources and label their consumers as "Referenced by tables (via merge/append/delegation)" instead of "Physical tables loaded".
• The lineage diagram colors non-loaded sources purple so power users can spot the hidden upstream layer at a glance.

No UI toggle — these are always included, matching the documenter's "reveal what PBIR hides" theme. Verified against an enterprise BigQuery dataset where ~19 previously-invisible *_src expressions now surface their GoogleBigQuery connections.

Live at https://jonathanjihwankim.github.io/pbip-documenter/ once GitHub Pages republishes. Please reopen if any non-loaded source still doesn't show up — thank you, this was a great catch!