Reject most shadow candidates in double before the exact test

[meshinspector-agent]

The shadow filter evaluated every candidate in exact arithmetic: the port to FastInt lost the bignum-to-double conversion the pre-filter needed. #6592 brings it back, so the pre-filter returns.

• up/uq are computed once per pair in double via toDouble; bp = dot( d, up ) and bq = dot( d, uq ), and a negative sign of either rejects the candidate before any bignum work.
• Error bound, which is what makes this safe: differences of points are below 2^31, so pp/pq are below 2^64 and the components of up/uq below 2^96. toDouble is correctly rounded, so each rounding costs 2^-53 of the running magnitude and the total error stays below 2^78 — a double below -2^78 is thus negative exactly as well. The tolerance is 2^84, 64x that margin; candidates in between fall through to the exact path unchanged.
• Base is agent/mlib-fastint-to-double (Add toDouble for the FastInt family #6592), not alpha-shape-triangle-shadow: toDouble is a hard compile dependency. Retarget once Add toDouble for the FastInt family #6592 lands on the branch.

The exact path itself is untouched — this only decides which candidates reach it.

Verification

Built in the ubuntu22 image mirroring build-test-ubuntu-x64.yml: clone_submodules_linux.sh --skip-prebuilt-thirdparty, prebuilt thirdparty linked as lib/include/share, Ninja Release, g++-12, -DMR_CXX_STANDARD=20. MRTest target and its dependency graph built clean, no new warnings.

MRTest --no-python-tests --gtest_also_run_disabled_tests --gtest_filter=*AlphaShape*:*FastInt*:*PrecisePredicates*:*InSphere* → 25/25 pass.

A/B on the same tree — DISABLED_AlphaShapeBench run with the pre-filter, then the file stashed and rebuilt incrementally so only this diff differs (same clouds, same binary otherwise):

cloud	triangle hash	shadowTests	exactShadowTests before → after	rejected
sphere 40k	17403291826481596167 unchanged	487416	487416 → 54994	88.7%
grid 40k	635805191690857954 unchanged	2058881	2058881 → 186448	90.9%
random 40k	11472133144093973002 unchanged	54564333	54564333 → 11466269	79.0%

consideredTris, touchableTris, inBallTests and shadowedNeis are identical on all three clouds in both runs, and so are the triangle hashes — the pre-filter is output-preserving, as its bound claims. inBallTests (1461420 / 3055058 / 23736199) also matches the filter-on column of #6590's table.

No timings here on purpose — this box is not a GitHub-hosted runner, so its milliseconds are not comparable to the reported set. The counters above are exact and platform-independent; the payoff in wall clock comes from the branch run.

Not exercised: the other 17 MRTest configs (MSVC, arm64, wasm, macOS) — this PR gets no checks of its own, the numbers come from the branch run after it lands. No UI surface (library-internal predicate, no MCP-exposed tool and no viewer widget), so nothing to drive and no screenshots; the three structurally different clouds x filter-on/off are the state coverage instead.

CI labels: none.

[meshinspector-agent]

Superseded by #6596: same step, rebased onto alpha-shape-triangle-shadow and computing pp/qq/pq as double dot products, so it no longer depends on toDouble. Force-push to agent/* is blocked, hence a new branch rather than a rebase in place.