Fix Rust build caching by using absolute paths

[michaelsproul]

We've had Lighthouse users on Discord reporting linking errors related to blst for a while, which would always be solved by cargo clean and a rebuild (suggesting a build caching issue). Today it happened to me and I decided to investigate.

I found that the build script's use of ../.. was causing the C compiler to reach out of the sandbox (OUT_DIR) given to it by Cargo, and into a path shared by all Cargo packages that shouldn't be modified from a build script. One of the compiler invocations it was making was:

running: "cc" "-O2" "-ffunction-sections" "-fdata-sections" "-fPIC" "-m64" "-Wall" "-Wextra" "-mno-avx" "-Wno-unused-command-line-argument" "-D__ADX__" "-o" "/home/michael/Programming/lighthouse3/target/release/build/blst-b4fcc12bd32c5abb/out/../../src/server.o" "-c" "../../src/server.c"

Note the ../.. in the -o argument that causes files to be written to target/release/build/src/server.o. If there are multiple versions of blst in the build cache (due to updates, etc), then they share these files, and I think that must confuse Cargo (unfortunately I can't describe an exact mechanism of action here). The Cargo docs have this to say:

Build scripts may save any output files in the directory specified in the OUT_DIR environment variable. Scripts should not modify any files outside of that directory.

In an attempt to fix it, I've switched to using absolute paths, and can confirm that the compiler invocations look more sensible now:

running: "cc" "-O2" "-ffunction-sections" "-fdata-sections" "-fPIC" "-m64" "-Wall" "-Wextra" "-mno-avx" "-Wno-unused-command-line-argument" "-D__ADX__" "-o" "/home/michael/Programming/blst/bindings/rust/target/release/build/blst-aa4544c2724aab9e/out/server.o" "-c" "/home/michael/Programming/blst/src/server.c"

And server.o ends up in OUT_DIR instead of its previous location. I hope this is the only build caching issue, and alleviates the problems our users have been facing 🤞

[michaelsproul]

I deleted the commented out code for bindgen, as that seems to be covered by build.sh now, but I'm happy to put it back if you'd like to keep it.

I contemplated deleting the override mechanism for local libblast.a and blst, but figured they were harmless because we never use them downstream, and they may be useful for you.

[dot-asm]

Cool! Thanks! I took liberty to align commentary with rustfmt.toml. And it also allowed to clean up CI scripts... In other words, merged. Thanks again:-)

[michaelsproul]

Awesome, thanks!