Forward object contains triangulation which uses omitted sources

I first computed source space for sample subject:
src = mne.setup_source_space(subject, spacing='ico5', subjects_dir=subjects_dir, add_dist=False)
Then forward solution:

model = mne.make_bem_model(subject=subject, ico=None,
                           subjects_dir=subjects_dir)
bem = mne.make_bem_solution(model)
fwd = mne.make_forward_solution(evoked.info, trans=trans, src=src, bem=bem,
                                meg=True, eeg=True, mindist=3.0, n_jobs=2)

Some source points were out of distance limits, so were deleted from the model:

9 source space points omitted because they are outside the inner skull surface.
114 source space points omitted because of the 3.0-mm distance limit.

But the forward object still contains triangulation which uses these omitted sources

print src[0]['nuse'], src[0]['nuse_tri']
print fwd['src'][0]['nuse'], fwd_fixed['src'][0]['nuse_tri']

10242 [20480]
10119 [20480]

After discussing this issue with @agramfort , a solution could be not to omit these sources, but to interpolate their lead fields using morph_data tool. I can do a PR for this, adding this option to
mne.make_forward_solution() function.

Start with mne.make_forward_solution() and inspect how the source-space nuse and nuse_tri fields are handled when distance limits omit points. Reproduce the discrepancy with the setup_source_space, make_bem_model, make_bem_solution, and forward-solution calls shown in the issue. The expected behavior and whether interpolation via morph_data is required still need to be agreed before implementation.