Reuse the objective constraint and order media bound assignment - #166

What this accomplishes

Two independent fixes in constraint and bound construction. Each ships a test that fails without it.

1. ObjConstPkg accumulated duplicate objective constraints

build_package added a new constraint on every call. Calling it twice — which gapfilling does when it tightens an objective threshold — left two objc constraints on the solver, with the stricter one silently binding. Nothing errored; the model was just more constrained than the caller asked for.

build_constraint now compares the linear coefficients of each existing objc constraint against the current objective and, on a match, updates that constraint's bounds in place and returns it. A genuinely different objective still gets its own constraint.

build_package also now returns the constraint. Previously it discarded the return value, so a caller had no handle to adjust the bounds later.

2. KBaseMediaPkg could pass through an inverted bound state

Exchange bounds were assigned lower-then-upper unconditionally:

reaction.lower_bound = -1 * self.parameters["default_uptake"]
reaction.upper_bound = self.parameters["default_excretion"]

When the media's uptake implies a lower bound above the reaction's current upper bound — a reaction already restricted to a negative upper bound — that order passes through a state where lb > ub. The upper bound is now raised first in that case.

KBaseMediaPkg also records the applied media on self.current_media, so packages built later can see what media is in force rather than re-deriving it.

Tests

tests/fbapkg/test_objconst_and_media_bounds.py — 5 cases on an inline two-reaction model: constraint reuse across repeated builds, the returned handle, a genuinely different objective still getting its own constraint, the inverted-bound ordering case, and current_media initialisation.

Verified on both sides: 3 of the 5 fail against the unpatched tree. The bound-ordering case was corrected after an initial version passed either way and therefore guarded nothing — it now sets upper_bound = -200 against a media uptake of 100, which is the condition the reorder exists for.

No fixture files, so these tests need no biochemistry database or commercial solver licence.

tests/fbapkg/test_objconst_and_media_bounds.py   5 passed
tests/                                          63 passed