Fix tableau coefficients using actual constructor values from old code

The previous extraction used wrong paper-format coefficients. This fix
uses the ACTUAL Alpha/Gamma/B/Bhat values from each constructor in the
old generic_rosenbrock.jl and properly applies _transformtab.

Key corrections:
- ROS2PR: gamma=0.228155 (was incorrectly 1.0)
- ROS2S: gamma=0.292893 (was incorrectly 1.7071)
- ROS3: correct A/C/b/btilde from Hairer/Wanner coefficients
- ROS3PR: gamma=0.788675 (was incorrectly 0.435866), now 3-stage
- Scholz4_7: correct 4-stage coefficients
- ROS34PW1a/1b: correct A/C matrices with proper nonzero entries
- ROS34PW2/PW3: minor precision corrections
- RosShamp4/Veldd4/Velds4/GRK4T/GRK4A/Ros4LStab: restored to 4-stage
  (were incorrectly collapsed to 3-stage)

Co-Authored-By: Chris Rackauckas <accounts@chrisrackauckas.com>

Author: ChrisRackauckas