[tclquadcode] New jump threading pass

[Kevin K. <kev...@gm...>]

I just committed and synced a new jump-threading pass to tcl-quadcode.

This replaces the old 'nodesplit' pass, which in a large procedure
could cause the entire optimization pipeline to rerun a great many
times for a large procedure, since it would thread only one basic
block at a time. (Moreover, it often faced a combinatorial explosion
of threading opportunities - many more than were justified by the
actual data types, because it did not detect converging pathways in
the splits.)

The new threading pass is the first one that I've done to require SSA
deconstruction and reconstruction. For the large number of changes
that the pass typically makes, it's faster and simpler simply to tear
things down and put them back together again.

With the new module in place, the 'renameTemps' pass, which was a
thorn in the side of the compiler on large procedures (owing to a
performance bug in Tcllib's 'struct::disjointset' module) can be
removed. In addition, of course, the old 'nodesplit.tcl' is also no
longer relevant.

Donal, feel free to ignore my earlier message about  addReference on
values combined with NEXIST IMPURE, and returnCode on values that are
FAIL+some other type. The revised jump threading separates those
combined types before they get to the offending operations. (I now
realize that the jump threading is *required* to compile [catch]
because the failure and success paths *must* be separated until the
FAIL is dismissed.)

Ad astra!

Kevin