Hmf

----------------------
**Enhancement**

- Normalised all <>_model properties to be actual classes, rather than either class or string.
- Added consistent checking of dictionaries for <>_params parameters.

----------------------
**Features**

- Now provides compatibility with Python 3.x. Support for 2.x will be removed in hmf v3.1 (whenever that comes).
- Complete overhaul of the caching system. Should be invisible to the user, but streamlines writing of framework
code considerably. Removes required manual specification of dependencies between quantities, and adds ability
to specify parameter kinds (model, param, res, option, switch).

**Bugfixes**

- Fixed bug in Caroll1992 GrowthFactor class which affected high-redshift growth factors (thanks to Olmo Piana).
- Fixed astropy dependency to be >= 1.1
- Fixed bug where Takahashi parameters were always passed through regardess of ``takahashi`` setting.
- Fixed small bug where the functional.get_label method returned differently ordered parameters because of dicts.
- Note that the fitting subpackage is temporarily unsupported and I discourage its use for the time being.

**Enhancement**

- Completely removes dependence on archaic pycamb package. Now supports natively supplied python interface to CAMB.
Install camb with ``pip install --egg camb``. This means that much more modern versions of CAMB can be used.
- Many new tests, to bring total coverage up to >80%, and continuous testing on Python 2.7, 3.5 and 3.6

--------------------------
**Bugfixes**

- Fixed bug in GrowthFactor which gave ripples in functions of z when a coarse grid was used. Thanks to mirochaj and
thomasguillet!

**Enhancments**

- Streamlined the caching framework a bit (removing cruft)
- Totally removed dependency on the Cache (super)class -- caching indexes now inherent to the called class.
- More robust parameter information based on introspection.

----------------------------

**Bugfixes**

- **IMPORTANT**: Fixed a bug in which updating the cosmology after creation did not update the transfer function.

-----------------------------
**Bugfixes**

- SharpK filter integrated over incorrect range of k, now fixed.

**Enhancments**

- WDM models now more consistent with MassFunction API.
- Better warning in HALOFIT module when derivatives don't work first time.

-------------------------

**Features**

- Added a bunch of information to each hmf_model, indicating simulation parameters from which the fit was derived.
- Added ``FromArray`` transfer model, allowing an array to be passed programmatically for ``k`` and ``T``.
- Added ``Carroll1992`` growth factor approximation model.

**Enhancments**

- how_big now gives the boxsize required to expect at least one halo above m in 95% of boxes.

**Bugfixes**

- Removed unnecessary multiplication by 1e6 in cosmo.py (thanks iw381)
- **IMPORTANT**: normalisation now calculated using convergent limits on k, rather than
user-supplied values.
- **IMPORTANT**: fixed bug in Bhattacharya fit, which was multiplying by an extra delta_c/sigma.
- fixed issue with ``nonlinear_mass`` raising exception when mass outside given mass range.