Tetrahedral remeshing: Give up on a flip as soon as its angles stop improving

[IasonManolas]

Summary

find_best_flip_to_improve_dh() searches, for each candidate flip vertex, over the cells the
flip would produce, tracking the worst dihedral-angle cosine seen so far. A vertex is only kept
if that worst angle is better than the edge's current worst angle. The search does not stop once
a candidate can no longer win: it keeps scanning the remaining cells and comparing angles that
can only make its worst-case value grow, never shrink.

Once the re running worst angle for a candidate vertexaches the edge's current worst angle, that
vertex is already disqualified (worse case can only stay the same or grow from there), so the
loop can stop scanning it right there instead of finishing the remaining cells. The same pattern
appears twice: once for each of the two symmetric flip-direction loops in this function.

The well-oriented/inverted-cells exception is preserved: for a candidate whose cells are inverted
(not well-oriented), the running worst angle is not a valid signal to reject on, so the early exit
only fires when is_sliver_well_oriented is true.

Testing

Verdict equivalence. The early exit stops scanning a vertex only after it is already
guaranteed to be rejected by the existing "is this vertex worth keeping" test, so it changes when
the loop stops, not what any candidate's outcome is. Full test suite (all executables in
test/Tetrahedral_remeshing) passes.

Byte-identical output, against plain cgal/main:

mesh	tets	byte-identical
bear	~40K	YES
bunny00	~225K	YES
101556.mesh (Thingi10K)	3,521,237	YES

Performance. perf stat -e instructions, Linux, Release, sequential, against plain
cgal/main.

mesh	iterations	cgal/main instr	this PR	Δ instr
bear	10	67.04 G	63.35 G	-5.52%
bunny00	10	148.80 G	139.60 G	-6.19%
101556.mesh	3	2441.29 G	2368.41 G	-2.99%

Release Management

• Affected package(s): Tetrahedral_remeshing
• License and copyright ownership: unchanged