Support bare :manifest() + chained field accessor for FILES - #272

Summary

$acme.files.:manifest():uuid() and $*.files.:manifest():uuid() now
work, literal-root and '*' alike -- driven directly by David's own
worked examples ("get uuids from all registrations," "get all acme
registration uuids"), meant to combine with SUBTRACT/INTERSECT
against another reference expression.

Background

Neither form worked before this PR. The first framing of the gap was
"does this need a . before the field accessor" -- traced live (parsing
both forms) and by code-reading before answering, since that turned out
to be the wrong axis. :first():fingerprint() (pointer + field accessor,
no dot, order-independent) already works today, but only because it
reads a field off exactly ONE already-selected entity. Bare :manifest()
does not select one entity -- it means "every entry in the manifest"
(Rule 1a/1b's own territory). Chaining a field accessor onto THAT asks
for something no code path did before: mapping a field over MANY entities
at once, regardless of dot placement.

Reference3.resolve_kind already classified this shape correctly as
METADATA_FIELD (confirmed live via terminal_functions), so the actual
gap was narrower than it first looked: query() needed to produce N
results instead of raising, and _extract_data() needed to read a field
per result instead of assuming exactly one.

What changed

• FilesReferenceFinder3._bare_manifest_field_call(name_one) -- new
  shared recognition helper: "name_one's path is a single, argument-less
  :manifest() call, with exactly one chained field-accessor function."
  Reused identically by query(), _query_star_traversal(), and
  _extract_data().
• query() (literal root): new branch reading the named-file's own
  get_manifest(), one ReferenceResult3 per entry.
• _query_star_traversal(): new branch reading
  files_root_manifest (the global arrivals ledger, Rule 1a's own
  resource) -- confirmed its entries use different key names than
  per-name manifest entries ("file_path"/"named_file_name", not
  "file") by reading FilesListener._prep_update() directly, rather
  than assuming the two manifests share a shape. Composes with
  :regex()-at-root_major.
• _extract_data(): widened the METADATA_FIELD field-lookup to also
  try the new helper, and to read from files_root_manifest instead of
  get_manifest(root_major) when root_major is the '*' token (not a
  real name).
• resolve()/resolve_from(): no changes needed -- confirmed by
  reading them first that the existing per-result _extract_data() loop
  already turns N results into N resolved values for free.

The pre-existing :name(...)+pointer, then :uuid() in name_three
pattern (narrow to one version, then read its field) is untouched and
stays equally valid -- David's own explicit confirmation, covered by a
dedicated test.

Deliberately not built: :all()/:flatten()/:groups() combined with a
chained field accessor -- partitioning semantics differ per marker, a
distinct question left on the bucket list.

Test plan

• TestBareManifestFieldAccessor (8 cases, literal root): one result
  per entry, resolved values (not whole entries), David's own exact
  syntax, a second field (:time()) to prove genericity, name_three
  rejection, more-than-one-chained-function rejection, non-field-
  accessor rejection, pre-existing name_three pattern unaffected.
• TestStarBareManifestFieldAccessor (4 cases, '*' root): same
  shape against a dedicated ledger fixture with the real
  "file_path"/"named_file_name" keys, :regex() composition,
  name_three rejection.
• tests/references/ full tree: 1555 passed.
• Full suite run twice for stability: 3148 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