Add calibration interface

[jamienoss]

Resolves #44 (partially, not inc #45)

Moved out of #47

The final design pattern is for the convert and fitting functionality to belong to the Transformer interface, and then the Calibrator interface takes these as config params for both input_transformer and output_transformer, along with this interface housing the runtime calibration procedure.

[codecov[bot]]

Codecov Report

Attention: Patch coverage is 97.70115% with 4 lines in your changes missing coverage. Please review.	Files	Patch %	Lines
evolver/calibration/interface.py	92.85%	4 Missing :warning:

:loudspeaker: Thoughts on this report? Let us know!

[jamienoss]

Update: We've discussed offline to adopt some of the design aspects I proposed a month or so ago, so this PR is still WIP.

The design points are:

• Remove is_calibrated semantics as this really isn't achievable - to really know whether calibration values are still accurate or not is not possible without running the calibration procedure to find out. Instead, we'll add timestamps to the calibration values and a mechanism to:
  • warn the user that these are old and could be inaccurate after some configurable time has elapsed
  • warn the user upon first use that they are using the shipped default values.
• Decompose "conversion" and "calibration" semantics, i.e., "conversion" is that between raw data and device units, e.g., voltage and temp in degrees C for a thermometer device. This yields a few advantages such as:
  • allows Output models to always have populated both raw and value fields, currently, this is not possible since value field(s) are those "calibrated" which may not have taken place yet. The Output classes also belongs to the hardware interface and not the calibrator interface and thus shouldn't be entangled. The conversion of raw to device output unit should belong to the device, not the calibrator.
  • allows for separate semantics & calibration implementations between absolute and relative calibrations. E.g., the linear relation between a thermometer's output voltage/resistance and temperature response is an absolute calibration that is unlikely to ever change, however, relative calibration between thermometers mounted on separate vials is vial dependent (thus as likely to change as smart sleeves are swapped - if at all) and really only requires just a calibration offset, not an offset and gradient as is needed for the V-temp relation (assuming it's linear).