Add support for BNGL models (cont.) - #508

Re-sync with PyBNF's sibling reader (pybnf/petab/_bngl.py, ADR-0026 / lanl/PyBNF#437):

- Block aliases: `begin molecules` / `begin species` / `begin rules` now open
  the same blocks as `molecule types` / `seed species` / `reaction rules`
  (_block_lines consults a per-canonical-name alias table), per the BNGL
  grammar reference (BioNetGen Perl2/; BNG_vscode_extension docs/bngl-grammar.md).
- Seed-species `$` clamp: `SeedSpeciesDefn = ["$"], Species, ...` -- the `$`
  fixed-concentration marker is stripped so `$counter() 10` enumerates the
  state variable `counter()`, keeping is_state_variable correct under the clamp.

Adds grammar-hardening tests (alias parsing, `$`-clamp stripping, no
cross-block shadowing, the is_state_variable seam) that double as the drift
anchor against PyBNF's reader. ruff check + format clean; 19 passed.

Refs: PEtab-dev/PEtab#436.

A trailing `\` (BNGL line continuation) splits one logical declaration across
physical lines. The block scanner processed physical lines, so a continued
parameter/function/observable was truncated at the `\` (e.g. `k = \` read as
the value `\`). Add _logical_lines() mirroring BNG2.pl's readFile
(Perl2/BNGModel.pm): strip the comment first, then while a line ends with `\`
drop it and concatenate the next comment-stripped physical line directly (no
space, so `1e\`+`3` -> `1e3`).

Surfaced by the bng_parity corpus (895 community BNGL models): 252 use line
continuation, incl. inside enumerated blocks (functions, observables,
parameters, seed species). Adds continuation + backslash-in-comment tests;
kept in sync with PyBNF's sibling reader (pybnf/petab/_bngl.py). ruff clean,
21 passed. Refs: PEtab-dev/PEtab#436.

BNGL declarations may carry a leading line label (LineLabel = {Digit}, WS |
Name, ":", [WS]): a legacy .net-style numeric index (`1 L0 1`) or a named label
(`CD14: CD14(...)`). The reader took the label as the entity -- the index as a
parameter name, the label as the seed species. Add _strip_line_label() and apply
it in the parameter and seed-species extractors (a valid BNGL identifier starts
with a letter, so a leading digit-run is unambiguously an index; a compartment
prefix carries `@`, so a bare `Name:` is unambiguously a label).

Surfaced by a writeModel-based differential over the bng_parity corpus (895
community models): 4 models disagreed with BNG2.pl's canonical parse (indexed
params/seed, labeled seed); after this fix, 0 -- parameters/observables/
functions/molecule-types/compartments all match BNG2.pl across the corpus.
Kept in sync with PyBNF's sibling reader. ruff clean, 24 passed.
Refs: PEtab-dev/PEtab#436.

Asserts parse_bngl enumerates the same model entities BNG2.pl does, over 21
curated public community BNGL models (RuleHub, BNGL-Models) under
tests/v1/bngl_corpus/. BNG2.pl's answers are cached in golden.json -- the entity
name sets it emits from `writeModel` (its canonical parse, no network
generation) -- so the test needs NO BNG2.pl and runs anywhere; it compares the
reader against the frozen oracle. Seed species are compared by molecule
composition to absorb BNG2.pl's pattern canonicalization (t vs t(), component
reordering, @compartment prefix vs suffix).

The models exercise every hardened reader path: line continuations, indexed and
labeled declarations, block aliases, the $ clamp, compartmental BNGL, energy
patterns, states/bonds, component reordering, bare-molecule seed species. The
golden is regenerated deliberately (needs BNG2.pl) via
`python tests/v1/test_bngl_corpus.py` and reviewed as a diff.

Mirrors PyBNF's live-BNG2.pl gate (lanl/PyBNF); validated there over the full
895-model bng_parity corpus (894/894 BNG2.pl-accepted models agree). ruff clean;
21 passed without BNG2.pl. Refs: PEtab-dev/PEtab#436.

…ies`)

The grammar doc lists `molecules` (for `molecule types`) and `rules` (for
`reaction rules`) as block aliases, but BNG2.pl 2.9.3 -- the reference this
reader targets -- REJECTS both ("Could not process block type 'molecules' /
'rules'"). Honoring them let the reader enumerate entities from a block BNG2.pl
refuses, i.e. accept models the reference rejects. Restrict _BLOCK_ALIASES to
`species` (for `seed species`), which BNG2.pl accepts and in fact emits as its
own canonical seed-species spelling. Verified empirically against BNG2.pl 2.9.3.

The corpus gate is unchanged (no fixture uses the dropped aliases; golden
regenerates identically). ruff clean; 24 passed. Refs: PEtab-dev/PEtab#436.

The 21 third-party .bngl fixtures under tests/v1/bngl_corpus/ are
unmodified (or, for Barua_2009, one-line-patched) copies of files already
published in RuleWorld/RuleHub and wshlavacek/BNGL-Models. Rather than
vendoring ~2600 lines of someone else's model text, fetch the same 21
files on demand from their pinned upstream commits, sha256-verified
against the exact bytes reviewed here.

golden.json and README.md stay committed -- they're this repo's own
oracle/test code, not sourced from anywhere upstream.

See scripts/fetch_bngl_corpus_demo.py for the fetcher.

Add tests/v1/fetch_bngl_corpus.py: fetches the 21 .bngl fixtures backing
tests/v1/test_bngl_corpus.py from their pinned upstream commits (RuleHub,
BNGL-Models) via jsdelivr's GitHub CDN, sha256-verified against the exact
bytes reviewed. No git/subprocess use, no execution of fetched content.

Wire it into CI as a step before the unit tests run, gitignore the fetched
.bngl files (golden.json/README.md stay tracked), and give
test_reader_matches_bng2_golden a real skip reason pointing at the fetch
script when the corpus hasn't been materialized yet.