Optimize AddMod for small reduced moduli

[perfloop-agent]

Summary

• Use fixed-width conditional reduction for reduced or near-reduced two- and three-word AddMod operands, plus direct scalar reduction for eligible one-word sums.
• Keep the wide and division-backed paths for inputs outside those bounds, preserve result aliases, and return zero before the small-modulus dispatch when the modulus is zero.
• Cover zero and one moduli, limb boundaries, carries, fallbacks, and aliases in the AddMod tests and fuzz seeds.

Benchmark

Workload: AddMod reduced operands / 64-bit modulus

Metric	Before	After	Change
ns/op	11.2	5.43	51.5% lower

Workload: AddMod reduced operands / 128-bit modulus

Metric	Before	After	Change
ns/op	23.11	5.875	74.6% lower

Workload: AddMod reduced operands / 192-bit modulus

Metric	Before	After	Change
ns/op	26.18	5.736	78.1% lower

Workload: AddMod unreduced operand / 64-bit modulus

Metric	Before	After	Change
ns/op	7.825	5.16	34.1% lower

Workload: AddMod operands at the modulus / 64-bit modulus

Metric	Before	After	Change
ns/op	25.09	4.911	80.4% lower

Workload: AddMod operands at the modulus / 128-bit modulus

Metric	Before	After	Change
ns/op	43.17	9.011	79.1% lower

Workload: AddMod operands at the modulus / 192-bit modulus

Metric	Before	After	Change
ns/op	66.23	11.67	82.4% lower

Test plan

• go test ./... -count=1
• golangci-lint run
• go test . -run - -fuzz FuzzAddModSmallModulus -fuzztime 20s
• go test . -run - -fuzz FuzzTernaryOperations -fuzztime 20s
• env GOARCH=386 go test -c -o /tmp/uint256.test-386 . && env GOARCH=ppc64 go test -c -o /tmp/uint256.test-ppc64 .

Checks: 5 passed.

---

Generated by Perfloop. Human sponsor: Tomás Senart. Measurements and checks.