Fix document/organization name filtering (client-side substring match)

[jason-sager]

Problem

itglue_list_documents (and itglue_list_organizations) advertise filter_name as a "partial match", but it doesn't filter at all — a name-filtered query returns the organization's entire document library.

Root cause: the IT Glue API's documents relationship endpoint does not support filter[name] (its only documented filter is filter[document_folder_id]), and where filter[name] exists on other resources it is exact whole-string match, never substring. The server sends filter[name]=<term> faithfully, but the API silently ignores the unrecognized param, so the two-call root + folder merge returns everything. This also contradicts the server's own instructions block, which says filters are exact-match.

There's also a secondary bug: list_documents reports the total as rootResult.total_count + folderResult.total_count, which double-counts.

Fix

Move name matching client-side, where substring matching is actually possible:

• Add ITGlueClient.getAll() — fully paginates an endpoint at page[size]=1000, with termination guards (hard page cap, empty-page break, non-advancing next_page break).
• itglue_list_documents: when filter_name is set, fetch all documents (root + folder), dedupe by id, match by case-insensitive substring, then paginate the filtered list client-side. filter_id (a genuine server-side exact match) stays on the wire.
• itglue_list_organizations: same treatment via a single getAll over /organizations; id/type/status filters stay server-side.
• Browse mode (no filter_name) is unchanged except the total is now the deduped row count instead of the double-counted sum.
• Corrected the misleading "partial match" / "exact match" wording in both schemas, both tool descriptions, the server instructions block, and the README so they match actual behavior.

Notes / trade-offs

• A name-filtered query now fetches the full list for the scope before matching (bounded: ~2 × ceil(docCount/1000) requests for documents; well within the 3000-req/5-min limit). Prefer filter_id when the exact ID is known.
• No new dependencies; no public tool signatures changed.

Tests

• New getAll unit tests (multi-page accumulation, all three guards, page[size]=1000).
• New search-mode tests for both tools (substring, case-insensitivity, dedup, client-side pagination).
• Updated the assertions that encoded the old on-the-wire filter[name] behavior and the double-counted total.
• npm test → 194 passing; tsc --noEmit clean.

🤖 Generated with Claude Code