Always round when calculating scaled integers

[pp-mo]

IMHO we should be rounding any time we convert a float value to a scaled integer.

While making a fix in #210, I scanned the whole of save_rules.py for "int(" and found several several other places in this code where rounding should probably be applied and is not :

• https://github.com/SciTools/iris-grib/blob/v0.15.1/iris_grib/_save_rules.py#L241
• https://github.com/SciTools/iris-grib/blob/v0.15.1/iris_grib/_save_rules.py#L355
• https://github.com/SciTools/iris-grib/blob/v0.15.1/iris_grib/_save_rules.py#L769
• https://github.com/SciTools/iris-grib/blob/v0.15.1/iris_grib/_save_rules.py#L1175
• https://github.com/SciTools/iris-grib/blob/v0.15.1/iris_grib/_save_rules.py#L1257

However, fixing the first of those breaks a test in Iris iris.tests.integration.format_interop.test_name_grib, which is also a CML-based one ...
which means we would need to release the new iris-grib, and only then fix Iris, as there is no sensible way to apply a tolerance to this test.

As #210 is to enable the version 0.16 release, I declined to take on the bigger issues at that time.
But maybe we should revisit + fix these other places.