An LTS release scheme, combined with encouraging libraries to maintain MSRV compatibility with LTS releases, could reduce this friction.

This actually sounds like a good idea. Currently crates are choosing their MSRV all over the place. If we just got a bit of alignment by calling every ~17th Rust release (roughly 2 years worth of releases) an “LTS” release, then crates could be encouraged to keep their MSRV compatible with that release.

But we also heard a consistent shape of gaps [in core]: many embedded and safety-critical projects want no_std-friendly building blocks (fixed-size collections, queues) and predictable math primitives, but do not want to rely on “just any” third-party crate at higher integrity levels.

I think some fixed-size collections and stuff like that would be super nice in core. Something with simple, predictable semantics, just like Vec has (i.e. no optimizations for certain usage patterns, like small string optimizations and that sort of stuff). With const generics working for integers, fixed size collections in core shouldn’t even be that hard (it’s certainly been done in many crates already).