Give a called analyzer its own _ROOT; look up bare words (4.1.1)

[ddehilster]

What

Two bugs reported while calling EnglishPhrases on a prose zone with callanalyzer (#739).

1. @NODES _ROOT and @PATH _ROOT matched nothing in a called analyzer. The called analyzer's tree root was the node it was given, under that node's own name (e.g. _PROSEZONE), sealed, and with siblings in the caller's tree. So:

• @NODES _ROOT / @PATH _ROOT selected nothing, and an analyzer whose passes all use @NODES _ROOT built nothing, silently;
• selects for deeper nodes never descended, since traversal only enters unsealed nodes;
• traverseSelect and pathSelect walk a root's right siblings, so a pass could run on the caller's nodes next to the zone.

Fix, in lite/callana.cpp only: for the length of the call the analyzer gets a stand-in _ROOT node (unsealed, no parent or siblings, spanning the node's text) that holds the node's children; they go back under the node afterwards. This leaves the ~20 select-matching sites in pat/rec/path/multi untouched, and makes a called analyzer select and traverse exactly as on a text of its own. pnroot() in a called analyzer is now that stand-in.

2. Bare words were never looked up in lazy *full dictionaries. CG::findWordConcept returned any word found in memory, even one with no attributes. Every attribute name is a dictionary word, so bare words are common, and a called analyzer shares the caller's KB: "see", "my" and "get" came out with no part of speech. A word in memory with no attributes now gets one lazy lookup (cached, so never repeated). A dict hit is merged into the word; a kbb-only hit is linked as meaning, as findFullWord already does.

Version bumped to 4.1.1.

Tests

The call-analyzer fixture's callee now uses @NODES _ROOT and @PATH _ROOT, and its caller makes "crimson" a bare word (as an attribute name) before calling. The previous binary fails it with the reported symptom (nothing built); this build passes.

Locally on Windows: all 15 fixture steps in build-windows.yml pass, including the rule-level debugger, and parse-en-us's final.tree is byte-identical to the golden copy. parse-en-us has no *full files, so the lookup change is covered by the fulldict-hang step and the fixture; it was not tested apart from fix 1.

Help: separate PR in VisualText/visualtext-files; merge it after this ships.

🤖 Generated with Claude Code