Pyrealm

- Replaced ConstrainedArray and masked arrays with input_bounds_checker and 'masking'
using np.nan. See notes in pyrealm/bounds_checker.py. This was revisited even before
release to remove built in masking completely and just provide some warnings on sane
ranges. A hard limit for temps < 25°C is imposed due to the behaviour of
calc_density_h2o.

- Fix for critical bug in mj calculation - using masked arrays is fragile, need to
consider this - and the constraint approach which generates masked inputs.

- Minor tweaks to utilities param classes
- Backtrack on constrained_arrays - unexpected issues with chained use. Currently just
turning off a single constraint.

- Refactor of constrained_array modules to use a class factory that acts as both a
constraint and a check on existing constraint types.
- Implementation of the utilities module, currently including some hygrometric
conversions and shared utility functions.
- Refactor of PModel and Iabs scaling
- Better **repr** and new summarize() functions in pmodel module.

- Refactor of the PModel to separate calc of gammastar etc, from the pmodel itself:
PModelEnvironment and PModel classes.
- Implementation of ConstrainedArray class to clip inputs to biologically meaningful
ranges and to identify that clipping has occurred. Particular issue with serious
numerical instability in calc_density_h2o, but now adopted a general solution to
clipping inputs.
- Expansion of PModel testing to include a global array giving a wider range of inputs
including edge cases.
- Created option for using a rootzonestress option (Rodolfo Nobrega)

- Restructure of requirements and install_requires for better pip install