[jhdark]

Adds TestTritiumBubbler4Physics: tests only, no source changes. Each test checks against hand-derived values:

• every output with distinct values on each input and parameter, so a swapped port or parameter fails;
• the four efficiency limits (0/1 for each efficiency), where the tritium distribution and the balance are exact;
• per-vial replacement timing (each vial restarts at its own time, the others are untouched) and a replacement at the final time;
• out-of-order replacement times are rejected;
• solver independence (EUF, RKCK54, ESDIRK43 give the same inventories);
• two instances in one simulation keep separate states and events;
• sim.reset() followed by a rerun reproduces the first run.

All pass on current master. I also checked that the tests detect injected bugs (swapped ports, wrong vial rate terms, wrong sample_out, resets hitting the wrong vial or all vials, per-vial times collapsed to one schedule, outputs in the wrong order); each one fails at least one new test, including one the existing suite missed (all vials resetting on vial 1's schedule).