Site response models + modification to Modifiable GMPE for AUS23

[mmpagani]

The AUS23 model uses a tabular GMM with fixed values of vs30. This does not permit computing hazard using spatially variable site conditions. This PR adds to OQ the linear and non-linear amplification terms used by GA to create the tabular GMMs. It also provides a function in the modifiable GMPE that permits using these two amplification terms.

We performed tests comparing the ground motion computed with the modifiable GMPE and the values available in the original tables distributed by GA (see https://ecat.ga.gov.au/geonetwork/srv/eng/catalog.search#/metadata/149177)

[g-weatherill]

The Hashash et al. (2020) and the Stewart et al. (2020) amplification models (and their epistemic uncertainties) were already implemented in the functions associated to the NGA East Base class (unless there has been an update to them). https://github.com/gem/oq-engine/blob/master/openquake/hazardlib/gsim/nga_east.py#L349 to https://github.com/gem/oq-engine/blob/master/openquake/hazardlib/gsim/nga_east.py#L517. If you are implementing the same functions again in separate files then I guess it would be useful to update the NGA East classes to call those functions from the separate files rather than have duplicate code.

[mmpagani]

@g-weatherill thanks. I completely missed them. I will check what you did back then and compare it to the code I prepared.

[g-weatherill]

@mmpagani I was initially thinking about just updating the NGA East class to refer to the functions in the files you are adding in this PR, but actually I'm seeing there will be some downstream consequences to doing that. The site amplification functions in the NGA East were implemented with their epistemic uncertainties, and these epistemic uncertainties are used as part of the ESHM20 logic tree for the craton region (https://github.com/gem/oq-engine/blob/master/openquake/hazardlib/gsim/eshm20_craton.py#L68) and for the USGS CEUS logic tree (https://github.com/gem/oq-engine/blob/master/openquake/hazardlib/gsim/usgs_ceus_2019.py#L253). So maybe it would be sensible to keep the two separate implementations initially and then expand the new implementation to include the epistemic uncertainties in amplification in a future PR where you can propagate the changes into the other affected classes.