What if any of DJB's choices made in eliminating
unnecessary sec-review complications applies to, or
is justified for digital signatures?
Or from the other side of the perspective:
re there design decisions made in Dilithium or
Falcon that can be justifiably deemed problematic
according to DJB's methodologies?
An analogous of decryption failure in Dilithium, would be the use of hints for compensating dropping over half of lower bits of public-key component $t$. qTESLA took the idea from Dilithium in round 2 and got itself broken.
As for Falcon, it once had some implementation bug in the Gaussian sampler, which amounts to a security concern and mis-estimate of instance parameters. The general design pattern of Falcon can be described as sophisticated, contrasting with many of DJB's work (e.g. Salsa/ChaCha-20, Curve25519, etc.) which can generally be described as simple and consistent.
- Is it possible and meritful to adapt some of the
NTRU Prime design decisions and strategies to
either of the lattice signatures from the finalists?
It should be clarified, that, it's not design elements that should be adapted to either of them. Arbitrarily applying an element over existing one provides little to no overall benefit.
Rather, it's the simplicity and consistency may be applied to their design, to eliminate exceptional elements.