VLS was originally developed for CLN (and primarily uses the internal interface of CLN to its signing component) so that remains the primary supported implementation. The CLN interface, although internal, goes through OS pipes to a separate signer process, and that was the easiest implementation to replace its signer at the time the project was started.

When LDK started exposing a formal documented signing interface, we also implemented that interface. It is in some ways drastically different from the CLN interface, so our support for LDK isn't quite as good as that for CLN yet (we do must of our testing against CLN). CLN has also started coordinating with us for some years already to somewhat improve their internal interface into something that actually works out better for a separated, higher-security signer with spending policies. For example, before, you would have to manually replace the CLN binary for its signer with our binary, but now, CLN has a command-line argument to do tell it to use a different signer binary for the signer process. LDK I believe still has some internal gaps where it expects to have the signing key in its process rather than calling into the formal published interface, especially for onchain funds (though don't hold me to that since I think LDK is actively fixing that).

LND still uses internal function calls to its signer (unlike CLN which does OS pipes into a separate process which is a lot easier to hack into without modifying CLN itself) that are undocumented (unlike LDK which has a formal documented signing intrface). I believe similar situation also with Eclair.