fix curv->CCS (3,0)

[slava77]

while cross-checking the results of #324 I found a typo; this PR fixes the curvilinear to CCS Jacobian element that couples q/p with 1/pt.
The impact of the mistake was a decrease in uncertainty of 1/pt by sinTheta^2 (cosLambda^2) in the input seed

http://uaf-10.t2.ucsd.edu/~slava77/figs/mic/mtv/10muHS_mk48b9f5b_sw5766e9f_vs_25bd4b9_v_mka3eddef-sw25bd4b9/

for built track initialStep:

orange is new/fixed

• the reconstructed track hit count is now roughly recovered
• the efficiency vs eta is now more evenly different from CMSSW
• the average efficiency vs pt did not change much

[slava77]

additionally, from standalone val in TTbar PU50 (left: older Cartesian -> CCS, middle: Curv-> CCS before this PR, right: fixed-up Curvilinear-> CCS with this PR)

nHits is back to the same distribution as expected; other distributions are mostly unchanged because once a few hits are added (as in the majority of the tracks) the fit mostly recovers the correct parameters.