Author SFPD, Groeipakket, and BFP Belgian source packages (ledger#69)

[MaxGhenis]

Summary

Completes the Belgian source-package set for ledger#69. PR #72 already landed the
six binding-selector streams (Statbel population + fiscal income, SPF Finances,
ONSS, ONEM/RVA, NBB) plus the EUROMOD-BE comparator and the NIS-2025 crosswalk.
This PR adds the three remaining publisher streams the issue body lists but that
were still missing, authored from primary government/statistical publishers only:

• SFPD legal pension caseload 2025 (sfpd-legal-pension-caseload-2025) — 4 facts.
  Federal Pensions Service (SFP/SFPD) figures via PensionStat.be, January 2025:
  all-schemes total 2,674,520 plus employee / self-employed / civil-servant counts.
• Opgroeien Groeipakket caseload 2025 (opgroeien-groeipakket-caseload-2025) —
  11 facts. Per-component child and family caseloads published by Opgroeien (the
  Flemish agency that administers the Groeipakket).
• BFP economic outlook June 2026 (bfp-economic-outlook-2026-06) — 5 facts,
  all typed assertion: source_projection. Headline projections from the Federal
  Planning Bureau's 12 June 2026 outlook (real GDP growth, CPI inflation, general
  government deficit as a share of GDP, for 2026 and 2031).

Each package is a source_package.yaml + declarative manifest + curated CSV under
packages/ and db/data/, with SHA-256-pinned artifacts and full lineage. All
three pass ledger validate-package and validate_facts; new regression tests in
tests/test_belgium_targets.py assert the exact published values.

Validation

• uv run ledger validate-package <id> --year <Y> — "valid": true, zero errors,
  for all three new packages (and all six pre-existing binding streams re-verified).
• uv run pytest tests/test_belgium_targets.py — 11 passed (7 existing + 4 new).
• uv run pytest tests/test_ledger_source_package.py — passed.
• uv run ledger --db ... load all — exit 0 with the new packages present.
• uv run ruff check ledger db tests — clean.
• tests/test_ledger_bundle.py bundle-coverage counts updated to the real
  end-to-end build numbers (fact_count 38939 -> 38959; source_count 25 -> 28;
  source_package_count 54 -> 57; period_count 13 -> 14; by_source_table 49 -> 52).

Data integrity — gaps recorded, never estimated

Every fact value traces verbatim to a published cell. Figures the publisher gives
only in rounded or paywalled form are recorded as gaps in the manifest notes, not
encoded as facts:

• Groeipakket basisbedrag child count: published only as ">1.6 million"
  (rounded lower bound) -> omitted from the child record set; the exact basisbedrag
  family count (930,010) is published and retained.
• SFPD pension expenditure: published only as a rounded "69 billion EUR" with no
  exact figure or reference year on the source page -> recorded as a gap.
• BFP employment and 2027-2030 GDP path: the page states "around 20 000" jobs in
  2026 and "nearly 230 000" over 2027-2031 (explicitly rounded), and the per-year GDP
  path is not on the press page -> recorded as gaps pending extraction from the full
  report PDF.

Faithful representation note: the SFPD scheme counts sum to more than the
all-schemes total because mixed-career pensioners are counted in each scheme they
draw from; the scheme rows are per-scheme recipient counts, not a partition. This is
documented in the package and asserted in the tests.

Two items for reviewer / lead attention

1. EUROMOD-BE comparator kept (open PR Remove EUROMOD comparator facts from Belgium targets #80 conflicts with this issue). PR Remove EUROMOD comparator facts from Belgium targets #80
   ("Remove EUROMOD comparator facts from Belgium targets") is currently open and
   unmerged. Removing that package would break test_belgium_targets.py:: test_belgium_euromod_comparator_has_source_urls_per_row and test_ledger_bundle.py
   (both currently green on main), and would violate issue Author ledger-be source packages for open Belgian aggregates with geography vintages #69 acceptance criterion Harden Arch target semantics for Microplex eCPS replacement #4,
   which requires the EUROMOD-BE comparator package to exist (feeding populace#264's
   reform_validation.json). Under the README/AGENTS.md boundary ("everything a
   publisher asserted is a fact"), JRC is the publisher of the EUROMOD-BE Country
   Report and the rows are typed entity_role: validation_comparator, never
   calibration targets. This PR therefore keeps the comparator and does not touch PR
   Remove EUROMOD comparator facts from Belgium targets #80's branch. Recommend closing Remove EUROMOD comparator facts from Belgium targets #80 as contradicting the issue's own acceptance
   criteria, but that is the lead's call.

2. Alias-registry edit. Registering the three packages required adding aliases to
   SOURCE_PACKAGE_ALIASES in ledger/source_package.py. That exact file is not in
   the ledger-source-ingestor allowed_paths list, though the edit is intrinsic to
   adding a source package (PR Add Belgium target source packages #72 did the same). Flagging for the human/judge review.

Refs #69. Unblocks #70 and populace #263/#264.