Add a test for what petab does with a derived BNGL parameter

[wshlavacek]

A BioNetGen parameters block can give a parameter an expression over other
parameters instead of a number, for example kon koff/Kd. Call that a derived
parameter. PyBNF reads BioNetGen models through the petab library when it works
with PEtab problems, and petab's BnglModel has two methods that report parameter
values.

The problem this test addresses is that nothing here pinned what those two
methods do with a derived parameter. The existing tests in TestNativeBnglModel
run against the exported demo model, whose three parameters are plain numbers,
so they say nothing about the case. That matters because a PEtab parameter table
may override or estimate the parameters a derived value is computed from. If
petab reported a value taken from the model file alone, that value would reach a
simulator as a constant, override the model's own expression, and stay at the
stale number while the parameters it depends on are fitted. PyBNF's own exporter
already refuses to give a derived parameter a nominal value, in
_numeric_nominal, so the two sides agree today, silently and by accident.

There is no version gate that would let us notice a change. PyBNF declares
petab>=0.9,<1 in pyproject.toml and in the setup action, uv.lock is not
committed, and continuous integration resolves that range on every run. Any
petab release below 1.0 is picked up automatically. The new test is the only
thing that would notice petab starting to report a value for a derived
parameter.

What the test asserts is the part that has to hold either way: a derived
parameter does not appear in the list of free parameter values, a plain number
still does, and asking for the derived parameter's value raises. It accepts
either ValueError or NotImplementedError, because released petab 0.9.0 raises
the second and the change proposed upstream in PEtab-dev/libpetab-python#517
raises the first.

Verification. The assertions were run against both versions of petab, selecting
each source tree with PYTHONPATH and printing the directory petab was actually
imported from, since a version string alone does not prove which copy is loaded.
Against released 0.9.0 the free values are {'koff': 0.1, 'Kd': 5.0} and the
value request raises NotImplementedError. Against the upstream branch the free
values are the same and the request raises ValueError.

One caveat worth recording: this test cannot be run in the local development
environment as it stands, because that environment holds petab 0.8.2 while
pyproject.toml asks for 0.9 or later. Six tests in TestNativeBnglModel already
fail there for the same reason, since petab 0.8.2 has no BioNetGen loader at
all. The cause is a stale uv.lock, which is not tracked in the repository, and
uv sync fixes it. Continuous integration installs the declared range and is
unaffected.