Research if the size difference makes it worth shipping 2 binary versions about seafowl HOT 1 CLOSED

Without LTO, release build (but with wasmtime since putting it behind a feature is difficult, as we'd also need to gate a few function-related structs and logical nodes behind #[cfg(feature)]), sizes in KB:

• No default features: 50000
• Just catalog-postgres: 53064
• Just frontend-postgres: 50704
• Just object-store-s3: 53928
• All features: 58444

With "fat" LTO, codegen-units=1

• No default features: 34460, stripped with --strip-all: 29412
• All features: 40024; stripped with --strip-all: 33040

(note this isn't compiled with -Os / -Oz)

So we get a 10-15% improvement by removing the PG catalog / S3 store / PG frontend features. I'll keep the features as disableable at compile time, but I don't think this improvement makes it worth shipping 2x as many binaries.

from seafowl.