Simplifies the data format for Heracles' results.

Multiple results are now produced as a dictionary with keys corresponding to the input fields:

{
    ("POS", "POS", 1, 1): ...,
    ("POS", "SHE", 1, 1): ...,
    ("SHE", "SHE", 1, 1): ...,
    ...
}

For the output from angular_power_spectra(), the individual entries are

• 1D arrays with shape (lmax + 1,) for TT,
• 2D arrays with shape (2, lmax + 1) for TE and TB,
• 2D arrays with shape (3, lmax + 1) for EE, BB, EB of an auto-correlation (where EB = BE), and
• 2D arrays with shape (4, lmax + 1) for EE, BB, EB, BE of a cross-correlation (where EB ≠ BE).

For the output from mixing_matrices(), the individual entries are (in the notation of Brown, Castro & Taylor 2005):

• the $M^{TT,TT}$ mixing matrix for scalar x scalar,

• the $M^{TE,TE} = M^{TB,TB}$ mixing matrix for scalar x spin,

• the stack of mixing matrices

  • $M^{EE,EE} = M^{BB,BB}$,
  • $M^{EE,BB} = M^{BB,EE}$,
  • $M^{EB,EB}$

  for spin x spin.

In practice, this more compact storage of results, with everything for one combination of fields in one array, is generally more convenient (see example notebook).

Internally, individual results are stored as the new heracles.Result class, with is a subclass of numpy's ndarray, and decays into a plain numpy array under almost all operations. Arrays of Result type have optional properties:

• result.ell -- None or the array of angular mode numbers for the result.
• result.axis -- None or the angular axis corresponding to ell.
• result.lower -- None or the lower bound of ell for binned results.
• result.upper -- None or the upper bound of ell for binned results.
• result.weight -- None or the weight in each bin for binned results.

This means that angular power spectra, mixing matrices, and their binned versions all use the exact same data type -- no more case distinction for structured arrays.

Consequently, it is no longer necessary to have separate functions for binning. Both spectra and mixing matrices can use heracles.binned().

The same is true for reading and writing results to file: both spectra and mixing matrices can be read using heracles.read() and written using heracles.write().

Finally, the new data type is flexible enough to support multiple ell axes for covariance matrices, as well as future use cases for bispectra and trispectra.

Closes: #207