CSM - refactor main code for clarity and to fix some bugs

- Add error messages when length_unit or time_unit is missing from AIM
- Try to be conservative with creating new variables just for the sake
of more concise dictionary access
- gm_units uses "g" for now, but make sure the following code will work
even when we potentially generalize beyond "g". Throw an error when
an expected gm_unit is encountered.
- Extend run_csm to save the capacity curves in a JSON file to support
post-processing.
- Refactor determine_response to start with opening inputs and checking
if they are valid.
- Change some variable names to be more descriptive and avoid using
upper case letters.
- Parse the demand, capacity, and damping models early on once
rather than parsing them once for each direction. We could also
initialize the models early, but I am not sure they'll behave well when
applied multiple times. It seemed safer to initialize them for each
direction separately for the time being. Further improvements seem like
low hanging fruits in that department.
- Improve how roof height is loaded. Further improvements are planned
by taking care of how roof height is saved in GeneralInformation.
- Add scaling of roof height by considering the workflow units and the
internal units of the CSM script
- Now that the capacity model is loaded ahead of time, loading the
default roof height from Hazus will actually work.
- Add a temporary fix to scale PEER timehistories assuming [g] units
rather than using the workflow units from AIM. This will be improved
later.
- Extend IMComputer part to also extract PGA and store it as PFA-0-1
- Add comments about the various unit conversions so that future devs
might find it easier to follow what is happening.
- Improve how accel_x and accel_y are loaded from response_accel
- Scale the resulting spectral accelerations back to the workflow units
before saving them in the EDP file.