[inputs.vsphere] Introduce use_int_samples to obtain backwards compatibility

[prydin]

Required for all PRs:

• Signed CLA.
• Associated README.md updated.
• Has appropriate unit tests.

TL;DR version:
In version 1.10, we changed the data type of all samples to floating point. This caused problems when sending data to existing InfluxDB databases. The use_int_samples forces the plugin to send int samples for everything except percentages.

Long version:
Under certain conditions, such as high load, vCenter can significantly delay sample delivery. When this happens, the plugin will see empty samples for the current time period. To solve this, we have to look back a certain number of sample periods to catch delayed data. Unfortunately, that data can come in at different time series granularity, so we apply a normalization process implemented in Endpoint.alignMetrics. This process consists of dividing the raw time series into buckets according to the desired sample period. If two or more samples end up in the same bucket, averaging occurs.

All vCenter metrics are integers, but when averaging occurs as described above, we need to use floating point math. Since the result of the normalization is a floating point vector, we changed the sample type to always be a float. Obviously, this was a mistake, since it breaks backwards compatibility.

The use_int_sample simply rounds output to the nearest integer and sends it as an integer type. Since the raw data was an integer, one could argue that this is just discarding false precision created in the averaging process. Either way, the rounding error should be negligible.