Patch for "no polynomial" case for azimuth FM rate mitigation & handling the burst without burst polygon

[seongsujeong]

This PR is an attempt to fix the rare case that the list pf polynomials (azimuth FM rate or doppler centroid) does not have any overlap with the sensing start / stop of the burst. In that case, the azimuth FM rate mismatch correction finds the nearest polynomials in azimuth time, but the indices for polynomials (index_start and index_end) has the same value. When that happens interpolator from InterpolatedUnivariateSpline cannot be generated.

When that happens, the index_start and index_end are adjusted so that anInterpolatedUnivariateSpline gets to have valid parameters to create the interpolator.
When using the generated interpolator, optional parameter ext=3 was set so that the interpolator returns the values at the left / right boundary values when the input parameter for the interpolation of out of range. This is equivalent to nearest neighbor in this case because the sensing start / stop is "out of range" from the list of polynomials' azimuth time.

Another issue that this PR handles is the when the burst does not have corresponding burst polygon. Such cases have been discovered as documented in #118. When that is the case, s1-reader will fill Sentinel1BurstSlc.border member with empty polygon. Likewise Sentinel1BurstSlc.center will be populated by an empty point geometry.

[seongsujeong]

@vbrancat @LiangJYu I think the PR is ready to review. I will submit a PR to COMPASS To change the code accordingly.

[vbrancat]

LGTM @seongsujeong. We reviewed this PR during our OPERA CSLC tag-up meeting. Please, add comments in the code on which data is affected by this fix. Thank you :)

[LiangJYu]

LGTM. Only a nit comment