FX.10 SRK bubble point lands on a wrong root near the critical region

lib/process: SRK methane/propane 40/60 at 30 bar flashes two-phase at 230 K, yet the bubble point at 30 bar is 269.8 K; the vapour-fraction flash inherits it. Likely wrong root near the critical region. Not fixed yet (docs/0.9-sessions/cape-open-09.md).

engineer-09 2026-11-05: Reproduced (SRK methane/propane 40/60, 30 bar): bubbleT 269.77 K while flashPT was already two-phase from about 215 K. Cause found: not the cubic's root choice. The saturation scan starts at 30 K, where the incipient vapour collapses onto the feed (the trivial solution); that collapsed composition was carried as the warm start to every later grid point, so the residual stayed the 'no second phase' flag (-1) up to 269.8 K, where the flag flipped to a real residual and the bisection took the flip as the root. Fixed in lib/process/thermo.js: a trivial warm start is thrown away and the Wilson seed tried; every sign change on the grid is a candidate and the first one that passes the stability condition (Michelsen tangent-plane test: stable just on the liquid side, unstable on the other) is returned (result.stable). Now 214.43 K; flashPVF(30 bar, V of the 230 K flash) gives back 230.000 K (was 269.77). Reference (measured, not typed from memory): May et al., J. Chem. Eng. Data 60 (2015) 3606-3620, doi 10.1021/acs.jced.5b00610, via NIST TRC ThermoML data sets 3-4, 243.62 K isotherm: x(C1) 0.3442 at 3909 kPa -> SRK (kij 0) 3943 kPa (+0.86 %); x 0.071 at 891 kPa -> 908 (+1.96 %). Tests: node tools/process-test/test.mjs 132/0 (10 new FX.10 checks), all 13 other process-engine suites one at a time, real page: node tools/process-test/browser-fx10.mjs 4/0 (/lab/cadcraft, the Flash and saturation calculator) and tools/capeopen-test/browser.mjs 13/0. Copies synced to cadcraft and bimcraft a Discord member/process.