Remove slow EnumerationCurves test

[d-torrance]

Runs out of memory on 32-bit systems and takes a long time (~25 s on my system) otherwise.

I asked Claude if we'd be losing anything by removing it, and this was its response (TL;DR: no):

Coverage Analysis: Removing EnumerationCurves TEST 1

Question

If TEST 1 (rationalCurve(5)) is removed, does any functionality of the package
lose test coverage?

Answer

No. All code paths exercised by TEST 1 are already covered by other tests,
primarily TEST 4.

What TEST 1 Uniquely Exercises

rationalCurve(5) triggers the d == 5 branch in fixedPoints
(EnumerationCurves.m2:122–179), which is the only place the helper functions
graph51–graph56 are called. Those six helpers are never reached by any
degree ≤ 4 computation.

Why TEST 4 Already Covers That Code Path

multipleCover(d) calls fixedPoints(1, d) internally. TEST 4 asserts
multipleCover(5) == 1/125, which calls fixedPoints(1, 5) and therefore
enters the d == 5 branch. With r = 1 (two torus weights, 0 and 1), the
inequality guards (k != h, h != m, m != n, etc.) are all satisfiable, so
graph51–graph56 are genuinely called and their results feed into the
asserted value.

Coverage Map After Removing TEST 1

Code path	Still covered by
rationalCurve (general)	TEST 0 — degrees 1–4
d == 5 branch in fixedPoints	TEST 4 via multipleCover(5)
graph51–graph56 helpers	TEST 4 via multipleCover(5)
contributionBundle, normalBundle	TESTSs 4–6
linesHypersurface	TEST 5
Internal helpers (torusList, fixedPoints directly)	TEST 6

What Is Lost

The only thing removed is the end-to-end verification of the value
229305888887648 (Katz's count of degree-5 rational curves on the general quintic
threefold). The algorithm producing it remains fully exercised through other
tests; only this specific published constant loses its direct assertion.

Closes: #4413