[cshenry]

What this accomplishes

Adds modelseedpy/fbapkg/objectivepkg.py and exports ObjectivePkg from modelseedpy.fbapkg.

It lets an objective be expressed as a string — MAX{(2)+rxn00001_c0|(3)rxn00002_c0} — parsed into a structured ObjectiveData, applied to a COBRApy model, cached by name, and restored later. A signed term weights one directional variable; an unsigned term weights both, with the reverse variable negated.

This is the mechanism MSGrowthPhenotypes uses to set a per-phenotype objective without the caller assembling optlang expressions by hand. It is a prerequisite for the phenotype-simulation work that follows, so it is submitted on its own first.

Class	Responsibility
ObjectiveTerm	one coefficient / variable / direction triple, with from_string and to_string
ObjectiveData	a sense plus a term list; parses and emits the MAX{…} / MIN{…} form and builds the COBRApy objective
ObjectivePkg	BaseFBAPkg subclass; build_package applies and caches an objective, restore_objective reinstates a cached one

This is a new file, so nothing existing imports it and the only change to current behaviour is one added export line.

Four defects fixed before submission

These were in the code as written and are fixed here rather than carried in:

1. Silent numeric corruption. ObjectiveTerm.from_string truncated the coefficient to a single character — float(array[0][1]) — so "(12)" parsed as 1 and "(2.5)" as 2. Now float(array[0][1:]).
2. String round-trip was lossy. to_string emitted a float repr, so a coefficient parsed from "(1)" came back as "(1.0)". Because build_package uses to_string() as both the cache key and the returned name, a caller could not restore_objective() with the string it had just passed in. Integral coefficients now emit without the trailing .0.
3. build_package returned the wrong value. It returned its objective_name parameter rather than self.objective_name, so it returned None whenever the name was derived from the objective string.
4. restore_objective discarded the cached original. It overwrote self.original_model_objective with the current objective before restoring, defeating the purpose of caching it in build_package.

Tests

tests/fbapkg/test_objectivepkg.py — 17 cases over an inline two-reaction model, covering term parsing (including the multi-digit and decimal regressions above), string round-trip for both senses, directional coefficient assignment, the missing-reaction warning, and the build / restore / cache paths.

No fixture files are used, so these tests do not depend on a biochemistry database or a commercial solver licence.

tests/fbapkg/test_objectivepkg.py  17 passed
tests/                             75 passed