[fft] Add fused graeffe engine primitive

[devin-ai-integration]

Summary

Adds E::graeffe(t, n) -> product to the engine contract: the size-n/2 even half of A(x) * A(-x) (one Graeffe/root-squaring step), semantically downsample(mul(t, negate_arg(t, n), n), n/2, false) but fused so engines can skip materializing the negated transform, the full-size pointwise product, and the separate downsample pass.

The fusion exploits that the product takes the same value A(w) A(-w) at the paired entries 2u, 2u+1 (which hold A at +-w in the bit-reversed convention):

• ntt: p[u] = t[2u] * t[2u+1] — one mult per output, no halving, no inv_rt twist.
• real: each A(w) A(-w) is already a plain spectrum value of the halved (real) result, so the pass computes (x + w y)(x - w y) per packed entry and repacks adjacent pairs with one retangle per output (mul's retangle degenerates since p0 == p1).
• split: inlines the negate shuffle (b[i] = t[i^1]) and the even-half average into mul_impl's conjugate-symmetry unpacking pass.
• crt delegates per prime; the algebra wrappers (matrix/trunc/*_stable) use the generic composition, which stays correct for the non-commutative case (the fused single-product form assumes commutativity).

Call sites switched: packed_bivariate::advance (the ps_compose Graeffe loop; negate_arg is still computed there since the caller needs the transform of Q(-x, y)) and the Q <- even part of Q(x) Q(-x) update in kth_term_of_rational_function. This saves O(n) work per level/iteration — a constant-factor win on the lower-order pointwise passes, not on the n log^2 n transform term.

New graeffe TEMPLATE_TEST_CASE covers all base engines at n = 2, 16, 64 against the slow product's even coefficients mod x^n - 1 (121 tests pass).

Link to Devin session: https://app.devin.ai/sessions/7d41aa62f31e4620807608a6e6afcc94
Requested by: @ecnerwala

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[github-actions]

GCC Code Coverage Report

📂 Overall coverage

Metric	Coverage
Lines	🟡 8211/10013 (82.0%)
Functions	🟢 1122/1227 (91.4%)
Branches	🟡 6418/8290 (77.4%)