Use portable builds by default

[antondlr]

Issue Addressed

Heavily reduces strain on CI infra during docker builds and releases
(docker builds: 9 -> 4* ; release builds 8 -> 4)
See #5457

Proposed Changes

Make blst-portable the default across the board and stop offering force-adx builds

Additional Info

In this PR, all current docker tags and direct release links** are preserved but lead to a portable build instead, which should work for everyone.

The idea is to release with these placeholders to try and assess how many downloads we still get on the to-be-deprecated release links.

Before we consider this, we should make absolutely sure that the performance hit, if present, is negligible on modern hardware.

(* 8 lighthouse and 1 lcli build --> 2 lighthouse and 2 lcli builds)
(** excluding the link for a non-portable windows build)

Summary of changes

• all lighthouse builds (both binaries and docker) use the portable blst library
• lcli is built through cross, and published as multiplatform image
• regular docker images now include spec-minimal

[antondlr]

Succesful CI runs (on subpar runners so timings aren't accurate):

release ci run release page
docker ci run lighthouse lcli

[antondlr]

Michael has already run some benchmarks

next step would be for us to dogfood this and deploy portable builds across the fleet.
it's now the default choice on our builder

[michaelsproul]

I'm ok with rolling it out. I was hesitant previously because the benchmarks showed a slight detriment, but we should gather some data on what impact this small slowdown has on the real workload

[paulhauner]

I'm onboard to try it out 👍

[antondlr]

I'm ok with rolling it out. I was hesitant previously because the benchmarks showed a slight detriment, but we should gather some data on what impact this small slowdown has on the real workload

We have portable builds running on our own fleet now, up to and including canaries, while mainnet runs the canonical non-portable release.
If we need more datapoints for the VC, I could update one or two mainnet VCs as they are quite homogenous, resource-wise. That should provide us with enough data.

[AgeManning]

I support this change also (under the believe that Michael's performance metrics show no real/significant performance degradation)

[antondlr]

closing in favour of #5614 (docker) and #5615 (binaries), which are cleaner, split-out versions of this PR.