As reported by Jonathan Kew on the list.

Testing with old Deva spec this reduces failures.
Test sequence: U+0915,U+094D,U+0930.

This reverts commit 0831061e.

Bengali is at *just* above 5% now.

Error down from 20% to 7%.

Yep, on Bengali now!

For dotted-circle independent clusters, Uniscribe does no Reph shaping
for the exact sequence Ra+Halant+25CC.  Which also is the only possible
sequence with 25CC at the end.

See comments.

Enable by setting envvar:

  HB_OT_INDIC_OPTIONS=uniscribe-bug-compatible

Plus, LeftMatra+Halant "feature".

As can be seen in: U+092B,U+093F,U+094D

Needed for upcoming changes to GSUB/GPOS mask matching.

This is to incorporate it into GSUB/GPOS processing.

Not that the code makes any difference since the presence of ZWJ itself
causes the ligature to fail to match anyway.

Another 1000 down!

Fixed for Deva cases with two full-form consonants.  Failures **way** down.
Not much left to go :-).

Hindi Wikipedia failures down to 6639 (0.938381%)!

Uniscribe doesn't do it, but we want to do as it gives the Reph the
opportunity to interact with the Matras.  Test with mangal for example.
Sequence: <0930,094d,0915,094b,094d>
In test suite already.