All per-pixel calibration data is now calculated in a separate structure
and then extracted. The signal for a flagged pixel is if *all* of the
*raw* values are 0, which necessitated a refactoring of the per-pixel
calculations and a mess of a function as it needs access to all of the
data in one pass.

Adding benchmarks exposed some pain points in the low-level API that
have since been fixed. The existing test data has also been moved into a
separate package so that the benchmarks can use a common data source.

The benchmarks have also exposed that using bitvec for the EEPROM
loading functions (bit twiddling and sign extension) is **much** slower
than the existing functions.

Merge branch 'benchmarks'

* benchmarks:
  Add pointer to example of low-level API usage.
  Show how FrameRate can be converted to a Duration
  Fix formatting in EEPROM benchmark
  MLX90641 support has been implemented for a while now
  Fix typo in CI build configuration
  Bump version to v0.3.0
  Update CI scripts for workspace and benchmarks
  Add benchmark for calibration processing
  Add benchmark for raw data to temperature calculations
  Always handle data resolution as Resolution, not u8
  Move register read/write implementation to traits
  Move read_ram into common and make it public.
  Split test data out into separate crate
  Convert existing structure into a workspace

`package`, not `project`.

The public API changed, specifically how Resolution is now used instead
of u8.

Both the two-pass and one pass functions are benchmarked.

CalibrationData::resolution() and MelexisCamera::resolution_correction()
now both take Resolution, and not the raw byte value. This was another
hurlde for users of the low-level API, as Resolution::to_raw() wasn't
public.

Read access is implemented through the existing FromI2C trait, while
writing is implemented through a new ToI2C trait.

This can be useful when using only the low-level API.

This makes it easier to write benchmarks using the example data.

Updated to remove deprecated functions, as well as fixing a couple of
new clippy lints.

Bugfix for MLX90640 IL mode.

This reverts commit 3c5140c4.

The main areas of improvement for this release are better no_std
support, improved support for non-default uses, and a slightly more
refined API in regards to constant values.