Improve DFP period functionality to allow for better sampling and ignoring period

[mdemoret-nv]

Description

Currently, the DFP pipeline simulates time by breaking incoming messages up by a specific period and processing each period independently. This makes it impossible to process all of the incoming data at once for batch mode with a single trained model.

This PR adds a few things:

• If DFPFileBatcherStage.period == None, then all messages will be processed in a single batch, instead of per period
• Fixes how periods were handled to work with counts
  • Before, "D" would work as expected but "5D" would not. This was due to using to_period
• The DFPFileBatcherStage.sampling_rate_s property was deprecated in favor of a more general sampling property
  • This property can support different values
    • If its a string, the value is interpreted as a frequency. The first row for each frequency will be taken
    • If its a value between [0,1), its a fraction. A percentage of rows will be taken
    • If its >=1, its a count. A random count of rows will be taken

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[mdemoret-nv]

@drobison00 Do any of the DFP modules need to be updated to match?

[mdemoret-nv]

/merge