Integrate harmonic mode selection and waveform plugin support - #5381

Allow mode_array entries to carry an eccentric harmonic index n as a
(l, m, n) 3-tuple alongside the usual (l, m) pairs, guarding non-
eccentric LAL-based approximants against it with a clear error instead
of a confusing unpack failure.

Add add_custom_waveform_modes() and the pycbc.waveform.fd_modes /
td_modes plugin entry-point groups, so external packages can register
mode-by-mode (rather than only summed) FD/TD waveform generators
through get_fd_waveform_modes/get_td_waveform_modes.

…ation

Targeted fixes for the two issues qlty attributed specifically to this
PR's diff (not pre-existing code):

- pycbc/waveform/__init__.py: add_custom_waveform_modes was imported
  but unused (F401); mark it as an explicit re-export
  (`as add_custom_waveform_modes`), matching the existing convention
  for this file's other re-exported names, without reordering the
  surrounding (pre-existing, never-isorted) import block -- doing so
  was tried and reverted, since it turns lines that were previously
  untouched by this branch into "changed" lines from git's point of
  view, which surfaces several unrelated pre-existing F401/F403
  findings in that block as if newly introduced by this PR.
- pycbc/waveform/plugin.py: sort the one import line inside the new
  add_custom_waveform_modes() function (I001).
- pycbc/waveform/waveform.py, pycbc/waveform/waveform_modes.py: apply
  ruff's quote/wrapping conventions only to the lines this PR actually
  added (mode_array validation in _check_lal_pars,
  get_hm_length_in_time, and get_imrphenomxh_modes).

Not fixed (left for a separate, maintainer-scoped decision): qlty
also reports these 3 files, plus __init__.py's whole import block, as
"unformatted"/"unsorted". Reproduced locally: running `ruff format`
on waveform.py alone changes 1000+ lines end to end (quote style,
import wrapping, blank lines) starting from line 39, i.e. the file has
never been run through ruff format -- none of that is attributable to
this PR's actual ~20-line diff. Reformatting whole files as a side
effect of this PR seemed like the wrong call to make unilaterally, so
qlty_check will still fail on those file-level findings.