Selection: report a bare word that names nothing (#124)

[vv137]

Fixes #124.

The bug

$ :color red not heliix        # typo for "helix"
Colored 602 atom(s) red
exit=0

A typo'd selector produced an empty selection, and not inverted that into
everything. Same figure-shaped failure as #116 and #121, from the same
function — and #121's end-of-input check couldn't see it, because a single
unknown word leaves no leftover token to trip on.

Why it wasn't a one-liner

The parser is handed one MolObject at a time while callers loop over
objects. 1crn parsed against 1ubq is a legitimate cross-object reference
that must yield empty without an error, so the loop can go on to reach
1crn. At single-object parse time a typo and a not-this-object name are
indistinguishable — which is exactly why the branch returned empty.

So the fix supplies the missing knowledge rather than guessing: a caller-side
knownObject lookup, bundled with the $name resolver into a
Selection::ParseEnv and built once by Application::selectionParseEnv() off
the whole ObjectStore. A word is a typo only when nothing anywhere answers
to it. With no lookup supplied the parser stays silent — a caller that can't
enumerate objects has no business accusing one, and that default is what keeps
the convenience overload from inventing false positives.

Covers the bare <obj> form, obj <name> — which was silent while the bare
spelling errored, the odd half of the pair — and <name>/*. Unknown selectors
are reported ahead of an unbalanced paren (the typo is a definite fact; the
paren is often its symptom), leftmost first.

A second bug fixed on the way

Object names containing - or . never resolved: readWord split my-model
into my, so :count my-model failed with unexpected '-'. The bare-object
branch now re-reads under object-name rules, as the digit-led path (1ubq)
already did — and a typo of one is reported whole (my-modelx) instead of as
its first fragment (my).

Verification

False positives were the main risk, so I went looking rather than assuming.
Confirmed still silent and correct: cross-object barewords under both scopes,
digit-led (1ubq) and hyphenated/dotted names, <obj>/(...), <obj>/*,
/obj/chain/resi/name, resolved and unresolved $name, disabled objects,
objects in other tabs, :foreach ... in <sel>, centroid()/pca() register
expressions, :label corner topleft = ..., and lib/residue_scan.mt
end-to-end. The 59-expression corpus stays byte-identical to the pre-#116
baseline. 317 selection assertions, 55 script assertions, all 6 suites green.

Scope, stated honestly

These still match nothing silently, so not <them> still inverts to
everything — they need rules of their own, not this lookup, and
docs/SELECTIONS.md now says so in a table rather than implying the class is
closed:

form	example
unresolved $name	not $typo
slash paths	not /nosuch/A/10
<obj>/(...)	not nosuchobj/(chain A)
unknown imgt region	not imgt nosuchregion
typo inside a value list	chain A heliix (read as a second chain id)

The last is the biggest remaining hole and the one closest to #124's shape —
:color red chain A heliix paints all of chain A and reports success. It can't
use knownObject (chain ids are opaque), but the imgt branch already voids
the whole term on an unknown region name, so there's precedent for stricter
list handling in the same function.

🤖 Generated with Claude Code