fix(event-display): keep jivexml backward track theta and charge

[SajalDevX]

JiveXML gives a track's polar angle as cotTheta, and JiveXMLLoader.getTracks turns it into theta with Math.atan(1 / cotTheta). For backward tracks (cotTheta < 0, eta < 0) that is negative, and the shift into (0, PI) only ran after track.dparams had been built. So every backward track was stored with:

• a negative dparams[3] (e.g. -PI/4 instead of 3PI/4 for cotTheta = -1)
• the sign of q/p in dparams[4] flipped, because |p| = pT / sin(theta) came out negative

track.eta was already right since it's computed after the shift.

JiveXML tracks don't carry a charge, so the q shown for a track, "Color by Charge" and the kinematics panel all read it from dparams[4] and showed the wrong charge for every backward track. The RK extrapolation for collections without polylines also takes its charge from dparams[4], so those tracks curved the wrong way.

This moves the theta < 0 shift to right after theta is computed, before momentum and dparams use it. I dropped the old "check if we need to flip phi" TODO with it: atan(1 / cotTheta) + PI is just the actual polar angle in (0, PI), so phi doesn't need to change.

Testing

Added src/tests/loaders/jivexml-loader.test.ts with one forward (cotTheta = 1) and one backward (cotTheta = -1) track, both with positive pt. It checks theta is PI/4 and 3PI/4, q/p is positive for both, and eta is +/-0.8814.

• Before the fix: the theta and charge tests fail (-0.785 vs 2.356, and -7.07e-5 vs +7.07e-5). The eta test passes.
• After the fix: all 3 pass. The rest of src/tests/loaders and src/tests/helpers still pass, including jivexml-menu-integration, rk-helper and phoenix-objects.
• Prettier check passes on both files. I did not run ESLint or the full yarn test:ci, and atlasesd-loader / jsroot-event-loader did not run locally because I did not have jsroot installed. Neither of those touches this code.

[github-actions]

🚀 Preview deployed: http://phoenix-pr-1041.surge.sh

Built from 90e6269.