Allow a field accessor alongside FILES' :all()/:groups() GROUP-mode pointer - #274

Summary

$acme.files.:all().:last():uuid() and $mixed.files.:groups().:last():uuid()
now resolve to one field value per (named-file, path) group, instead of
raising.

Background

The bucket-list entry read as a single, still-open question -- "GROUP
modes combined with :manifest()/a field-accessor... not yet built for
FILES." Traced query()'s own GROUP-mode rejection and the reduction
logic just below it before touching anything, rather than assuming the
whole thing needed new machinery:

• The reduction logic (if pointers: if partitioned: ...) already
  independently reduces each distinct file_home's own candidates via
  the pointer, unconditionally -- nothing about it depends on whether
  :manifest()/a field accessor is also present.
  ReferenceResults3.ambiguous_content_read was already correctly
  written to flag only has_manifest, never has_field_function -- the
  Rule-1-vs-Rule-3 (whole-resource vs. poolable-field) distinction this
  fix depends on already existed, just wasn't consistently applied.
• The rejection's own comment claimed it "mirrors"
  ResultsReferenceFinder3's own :all()-grouping restriction -- but
  that was citing RESULTS' original, since-superseded blanket
  rejection. RESULTS' current _star_group_and_reduce() already allows a
  field accessor through here (e.g. ":all():last().invoices:uuid()"
  resolves fine there, confirmed live before that code was even
  written), only rejecting a whole-resource content accessor. FILES had
  simply never been updated to match -- a real, confirmable staleness,
  not a guess.
• _extract_data() needed no changes at all -- :all()/:groups()
  bare occupies name_one's entire content by construction, so the
  trailing field accessor has nowhere to land except name_three, the
  same already-tested extraction path every other name_three field
  accessor already uses.

What changed

Narrowed query()'s GROUP-mode rejection from pointers and (has_manifest or has_field_function) to pointers and has_manifest --
:manifest() (whole-resource) stays rejected here, unchanged, for the
same Rule-1 reasoning it always had; a field accessor is no longer caught
by the same check. That is the entire fix.

Deliberately out of scope: '*'-traversal's own, separate and much
broader field-accessor rejection in _query_star_traversal() -- a
bigger, still-open gap, unrelated to this fix's partitioned-plus-
pointers scope. Confirmed its own existing tests still pass unchanged.

Test plan

• Converted the two existing tests that asserted the old rejection
  into positive assertions -- one correct uuid per group, for both
  :all() (ALPHA_MANIFEST's two file_home groups) and :groups()
  (MIXED_MANIFEST's two different-depth groups).
• Their :manifest()-combined sibling tests (asserting the
  still-correct rejection) left unchanged.
• tests/references/ full tree: 1560 passed.
• Full suite run twice for stability: 3153 passed, 11 failed both
  times (identical) -- the known, unrelated SFTP/S3 failures
  requiring unset env vars (issue test_nos_server_config_1 crashes with a raw ValueError when SFTP_* env vars are unset, instead of failing clearly #216).

https://claude.ai/code/session_013gPjejPWvbWtjz2dw9h14M