Close the proton leak when fitting the measured rates

[edkerk]

With the measured oxygen fitted, the proton leak (r_2129) carried most of the
respiration at high C/N. It has no enzyme and no bound, so it is a free oxygen sink.

	CN4	CN22	CN38	CN75	hGR
leak, mmol/gDW/h	0	4.9	18.6	37.7	0
P/O	1.22	0.95	0.57	0.19	1.20

The published run had no leak flux (and under-predicted oxygen: 1.10 against 6.66
measured at CN75). The cause is that the built models can carry no maintenance ATP: the
most non-growth ATP they can support is 0 in every condition, and 5.3, 8.1, 14.8 and 4.9
at CN22, CN38, CN75 and hGR with every enzyme cap removed, which matches the sampled
conventional model (4.7, 9.0, 15.5, 5.1). The rate fit raises each cap only as far as the
rates need, so the oxygen goes through the free leak.

Change

--block RXN ... closes the named reactions (bounds 0, 0) before anything else is
applied; an unknown ID is an error. The README recipe now closes r_2129.

Result

With the leak closed all five conditions build, P/O is 1.10 to 1.20 everywhere, and the
non-growth ATP shows up as maintenance: 4.9 at CN38 and 8.2 at CN75. The exchange fit is
unchanged (mean residual 0.127 against 0.123). The protein pool used rises where the leak
had absorbed oxygen (CN22 57 to 74, CN38 62 to 79, CN75 64 to 102 mg/gDW). TCA median
capacity usage at CN38 and CN75 falls (16 to 4 and 44 to 6%).

fit_ngam is still uninformative under --fit-rates (0, error inf); the maintenance
comes out of the solution instead.

Tests

Two on a small model and two on the real model (blocking, and an unknown reaction), and an
integration build of CN22 with the leak closed: zero leak flux, growth at the dilution rate,
P/O above 0.9. 197 tests pass on Gurobi.