Allow a field accessor alongside FILES' '*'-traversal pointer - #275

Summary

$*.files.*.:last():uuid() (POOL), $*.files.:all().:last():uuid()
(GROUP, one level), and $*.files.:groups().:last():uuid()/
$*.files.:flatten().:last():uuid() (any depth) all now work, instead
of raising -- the '*'-traversal peer of the literal-root GROUP-mode fix
from #274, one level broader (every traversal shape, not just the
literal-root one).

Background

_query_star_traversal()'s shared name_three-processing tail had a
single blanket check rejecting :manifest() and any field accessor
together, for every traversal shape -- broader than the literal-root
check #274 narrowed. Traced the reduction logic just below it before
touching anything: it already independently reduces each file_home's
own candidates (GROUP) or pools to one overall candidate (POOL)
regardless of which accessor rides alongside the pointer -- nothing
there needed to change.

The one new question this raised beyond the literal-root case:
_extract_data()'s Star3-aware branch (added in #272, reading the
global arrivals ledger since result.uuid alone does not say which
named-file a '*'-traversal candidate came from) needed to generalize
from the name_one-sourced field_call it was built for to a
name_three-sourced one too. Confirmed it already does -- that branch
lives inside the shared if field_call is not None: block -- so no code
change was needed there, only test fixtures needed a ledger they never
previously required (the shapes reaching that branch were always
rejected before).

:manifest() (whole-resource) stays rejected everywhere in this method,
deliberately -- GROUP mode can produce more than one reduced candidate,
and this method sets no ambiguous_content_read flag to catch resolving
several whole entries at once (Rule 1).

Also corrected a stale docstring/comment claiming the restriction
existed because _extract_data() "cannot know" which manifest to read --
no longer true (the Star3-aware ledger branch already solves that), and
Rule 1 was the real, still-current reason even before that.

What changed

Narrowed the shared unsupported check from has_manifest OR field_ accessor to has_manifest only, mirroring #274's exact fix.

Test plan

• The one existing test asserting the old field-accessor rejection
  (POOL mode) converted to a positive assertion.
• Three new tests added for shapes that were never testable before:
  GROUP one-level, GROUP any-depth, POOL any-depth.
• STAR_LEDGER/FLATTEN_LEDGER fixtures added -- every other
  star-traversal test in this file still does not need one.
• :manifest()-combined tests (still-correct rejection) left
  unchanged in every mode.
• tests/references/ full tree: 1563 passed.
• Full suite run twice for stability: 3156 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