Use proper rounding when converting blocking terms to avoid rounding
issues.  This could cause some (very rare) off-by-one bugs when
incorrectly truncating an objective result very close to the next
largest integer.

This can reveal bugs in constraints that accidentally include Ti.

Add a bridge to the GLPK library. Availability is auto-discovered by
the build system. Most of the code is solver-agnostic.

Interestingly, this shows that generating huge LPs is not necessarily
any faster than creating and solving many small LPs. Further
investigation required.

This patch introduces the capability to generate
one huge LP for an entire task set, instead of
generating many smaller LPs (one for each task).

It's not clear yet that this is actually any faster.

These files implement the generation and evaluation of linear programs
that bound maximum s-aware pi-blocking under the DPCP and the DFLP.