Improve the validation of IMTs

[Michele Simionato]

The tricky case is when the IMTs are read from the risk models. One must treat correctly the duplicates. Moreover there is an issue when computing the GMFs with correlation: only PGA and SA are supported. This should be checked before starting the computation.

[Michele Simionato (michele-simionato)]

For instance in oq-lite there is still a problem if inside `intensity_measure_types_and_levels` you type "SA(0.750)", because later on the IMT will be referred as "SA(0.75)"

[Michele Simionato (michele-simionato)]

One must make sure that all IMTs are supported by all GSIMs. If not, an error must be raised in the pre_execute phase. The user is then expected to remove the unsupported IMTs (typically in the vulnerability model files), or to replace them with supported IMTs. Or, she can change the GSIMs.

[Michele Simionato (michele-simionato)]

https://github.com/gem/oq-risklib/pull/265