Fix AuCPake benchmark

During writing, I made the mistake and assumed that our STM32F407
benchmark would run at 84 MHz. However, we opted to choose the
default clocking profile instead. After assuring this for our own
benchmark, I forgot to update the AuCPake cycle count.

We should take the 16 MHz cc measurement though, because it better
resembles our own benchmarking setup.

Thanks to Matthias for finding this mistake.

results.tex

@@ -49,7 +49,7 @@ none of our implementations exceed the performance of Curve25519.
     % Their code is at https://github.com/armfazh/rfc7748_precomputed
     % So in <https://hyperelliptic.org/tanja/lc17/ascrypto/day1/slides/lopez-software.pdf> Lopez reports 127kcc, using MULX+ADCX/ADOX, but Haswell does not support ADX (only BMI2).
     Curve25519~(\cite{FA17}) & -- & -- & -- & $907\,240$\tmark[a]\\
-    Curve25519~(\cite{HL19}) & -- & -- & -- & $626\,719$\tmark[a]\\
+    Curve25519~(\cite{HL19}) & -- & -- & -- & $625\,358$\tmark[a]\\
     Curve13318~(\textbf{this work}) & $389\,546$\tmark[b] & $382\,966$\tmark[b] & $204\,643$\tmark[b] & $1\,797\,451$\tmark[b] \\
     Ed25519 verify & $221\,988$\tmark[d] & $206\,080$\tmark[d] & $184\,052$\tmark[d] & -- \\
     % SB: SUPERCOP h6sandy