Better reliability of `lazy_read_iiasa()`

[danielhuppmann]

#696 added a top-level function lazy_read_iiasa() which tries to load from file before querying the database. One problem with the current implementation is that the file does not "remember" the filters applied for querying an existing file.

For example, if you

df = lazy_read_iiasa("some/file.csv", TEST_API, model="model_a")

and later

df = lazy_read_iiasa("some/file.csv", TEST_API, model="model_b")

you will get a different result for the second call depending on whether any data in model_b changed in the database in the meantime.

I think that @phackstock implemented some smart solutions for such a problem in the downscaling-package...