*) Dedicated code paths for real and complex arithmetic
*) Improvements to reconstruction code
*) Speed up by 30% (check!)
*) New inner_bound_type parameter:
   - REGULAR (default); requires x_i = 0
   - ZERO; requires x_i /= 0
*) New rotation_method parameter:
   - DOPPLER
   - TRAD
   - NULL (default)

TODO
----

*) Fix up treatment of banded matrices
*) Fix grid generation to account for rot stuff
 - no account so far is made
 - current approach for resamp_dispersion is not robust
   - for each point in the grid, we work out the largest possible radial wavenumber that can arise in the frequency interval [omega_min,omega_max]
   - however, no account is made of the fact that 
     (i) the radial wavenumber depends on omega_c(omega), and not omega;
     (ii) the radial wavenumber depends on lambda(omega_c(omega)), and not l
   - with (i), we could just map [omega_min,omega_max] into [omega_c_min,omega_c_max] (since the mapping is monotonic)
   - but with (ii), we cannot
   - however, if we approximate lambda via a linear fit in w^2, then it just might work...
   - ...nah, that's naff
   - how about separate codepaths to work out the local min/max wavenumber
     - "fast" for DOPP, NULL rot_t types -- as before
     - "slow" for TRAD
     - is this even required, given the small amount of time "slow" takes?
     - yes, it *is* slow!
   - can we do root finding to look for an extremum?
*) Fix cutoff freq dependency on rot [done; simple warning]
*) Add freq_frame parameter [done]
*) Rename some parameters
   - rotation_type -> rot_method [not doing]
   - ivp_solver_type -> ivp_solver
*) Add NULL rotationt type [done]
*) Fix up gyre_mode_funcs for rotation [done]
*) Implement freq_frame functionality
*) Fix bugs stopping doppler test from working [done -- the issue was with the inlist]
*) Index/offset grid functionality
*) Get build_poly stuff working from command line arguments

GYRE monolithic / lib
---------------------

*) Why
  - lots of code shared between ad, nad and mesa
  - why multiple codes?
*) Shared code
  - initialization
  - parameter selection
  - build grid/build scan
  - mode processing
*) Do we need to change our overall computational approach?
  - daisy-chain the ad/nad find (but this would involve a rewrite of prox search)
