non-closure phase bias estimation (#765)

This PR adds the bias time-series estimation introduced by the non-closure phase, as described in Zheng et al. (2022, TGRS), via `--action quick_estimate / estimate` options.

Detailed changes are as below:

+ `utils.isce_utils.py`
   - add `guassian_kernel()` and `convolve()` for the complex interferogram filtering.
   - add `estimate_coherence()` and `unwrap_snaphu()` based on `isce2/topsStack` to re-unwrap the closure phase interferogram, to mitigate the PU error during closure phase bias estimation, including notes on SNAPHU from Piyush.

+ `objects.stack.ifgramStack`:
   - move `split2boxes()` from `ifgram_inversion.py` to here for more convenient re-use, add `dim0_size` option for more flexibility
   - add `get_closure_phase_index()` to return the indices of interferograms that forms the given connection level of closure loop
   - add `get_sequential_closure_phase()` for the 3D CP calculation and its temporal average.

+ `ifgram_inversion.estimate_timeseries()`: support `inv_quality_name = no` to turn OFF the calculation.

+ closure_phase_bias.py:
   - add `--action mask` for average CP and mask calculation, by refactoring the existing script.
   - add `compute_unwrap_closure_phase()` to compute the wrapped sequential closure phase, unwrap them using isce2 modules, and then calculate the cumulative unwrapped seq closure phase time-series via `cum_seq_unw_closure_phase_timeseries()`, with parallel support and auto-skip.
   - add `estimate_wratio()` and `estimate_wratio_all()` for the `W_r` estimation, a.k.a., the weighted closure phase history.
   - add `bandwidth2num_ifgram()`
   - add `estimate_bias_timeseries_approx()` for `--action quick_estimate` 
   - add `estimate_bias_timeseries()` for `--action estimate` with parallel support using dask.
   - 2X speedup of `estimate_bias_timeseries()` via `ifgram_inversion.estimate_timeseries()` with `scipy.linalg`, instead of `np.linalg.pinv`, and auto-skip the zero/nan values in part of the interferograms among the stack.
   - 2X speed up via common design matrix operation, same as in `ifgram_inversion.py`

Co-authored-by: Zhang Yunjun <yunjunzgeo@gmail.com>