Revise mnewton for greater simplicity and speed.

Revise mnewton for greater simplicity and speed.

In share/mnewton/mnewton.mac:

 * Call linsolve_by_lu directly instead of making a detour through
   augcoefmatrix -- we already have the coefficient matrix and
   right-hand side in hand, we don't need augcoefmatrix to undo the mess
   created by multiplying stuff together.

 * Calling linsolve_by_lu directly obviates the need for the list of
   symbols h. This also simplifies the list of increments.

 * Cut out keepfloat and numer. Those are not needed, and numer in
   particular causes expressions to become much larger, due to the
   presence of float exponents (1.5 instead of 3/2, etc.).
   The test cases in rtest_mnewton.mac now run much faster since the
   expressions involved are much smaller.

 * Additional debugging output: DF, Solutions, FuncList.

In share/mnewton/rtest_mnewton.mac:

 * Avoid kill(all) at start and end (kill only symbols used in
   equations), and load mnewton only if not already present.

 * Cut out now-irrelevant assignment to algexact and unused assignment
   to %start.

 * Move fpprec assignment to the place it is used.