fix: regenerate the bundled fixture against the sparsified corpus (#18)

The fixture predated three shipped corpus changes at once: no J rows in
summary_categories (it was built before the crosswalk completion), no
representation.parquet or code_set.parquet, and a still-dense wide era.
Every test in this package and in cog-api runs against it, so both suites
were green against a corpus that no longer exists. This is #18's stated
prerequisite; it proves nothing about production until it lands.

Regenerated from the publish tree at pipeline_commit 83f9715 (schema v6,
built 2026-07-29). FY2011 goes from 2,864,212 rows to 496,004 -- 82.7% of
the old partition was explicit zeros -- and the fixture now ships all ten
publish-tree metadata tables rather than six. The generator's file list is
a single constant now, so the copy step and the manifest step cannot drift.

Three test repairs, each a real consequence of sparsification rather than
a number to bump:

  test-categories.R          "assistance" joined the spending subtype
                             vocabulary with the J-prefix codes.

  test-spending.R            The harmonization block counts rows that
                             exist. Broward's E21/F21/G21 were zero-pads
                             and are gone, so the anchor moves to FL state,
                             whose three NA-mapped rows carry $2.83B --
                             the amount accounting was previously asserted
                             only against 0 and could not have caught a
                             bug. Broward keeps a test of its own, now
                             asserting the zero-pads are absent.

  test-expenditure-concept.R Coverage-gap suggestions are presence-based.
                             AL state's only FY2011 B47 cell was an
                             explicit zero, so ig_federal_b47_wide stopped
                             being a candidate there; FL state carries a
                             real amount, so the counterpart guard is
                             exercised against a suggestion that fires.

test-fixture-vintage.R pins the structural facts that separate this vintage
from its predecessor -- the ten metadata tables, the dense/sparse
representation contract, zero explicit zeros in FY2011, code_set coverage,
and J19's category. Checked against the old fixture: FY2011 carried
2,368,208 explicit zeros, so the assertion discriminates rather than
merely passing.

Suites: uscogdata 594 pass / 0 fail / 6 skip (was 576/0/6).
cog-api 357 pass / 0 fail / 8 skip against the regenerated fixture,
unchanged from its baseline.

tests/testthat/test-expenditure-concept.R

@@ -298,8 +298,16 @@ test_that("a mis-scoped cog_spending() call never attaches an M/L counterpart to
   # (M47/M94, same suffixes) -- a coincidence of reused digits, not a real
   # Direct/Total pairing. The flow-family gate in
   # .attach_ig_counterparts() must keep ig_recipe_id NULL here.
+  #
+  # Anchored on FL state government, not AL. Coverage is presence-based: a
+  # recipe is only suggested when its component codes have rows for the
+  # requested government-year. AL state's only FY2011 B47 cell was an
+  # explicit zero, which the corpus no longer stores after sparsification
+  # (SB194, cog_pipeline#64), so the recipe stopped being a candidate there.
+  # FL state carries a real FY2011 B47 amount, so this exercises the guard
+  # against a suggestion that genuinely fires.
   r <- suppressMessages(
-    cog_spending("010000226085", years = c(2005, 2011), category = "IG Federal")
+    cog_spending("120000226351", years = c(2005, 2011), category = "IG Federal")
   )
   sugg <- attr(r, "provenance")$suggestions
   expect_gt(length(sugg), 0L)