Restrict CI on forks.

[nnethercote]

• Don't do Windows CI on forks. (We already don't do Linux CI on forks.)

• Don't do container image CI on forks. (We shouldn't be pushing container images to ghcr.io from forks!)

• Remove the if condition in deploy_guide.yml. It has no effect because there is an on: push: branches: - main condition higher up doing the same thing (and in the more typical way).

[nnethercote]

The CI failure is the following. I saw the same thing in #267, so I think it's unrelated to this PR.

error: failed to run custom build command for `vecadd v0.1.0 (/__w/rust-cuda/rust-cuda/examples/cuda/vecadd)`
note: To improve backtraces for build dependencies, set the CARGO_PROFILE_DEV_BUILD_OVERRIDE_DEBUG=true environment variable to enable debug information generation.

Caused by:
  process didn't exit successfully: `/__w/rust-cuda/rust-cuda/target/debug/build/vecadd-78141776eacf0aee/build-script-build` (exit status: 101)
  --- stdout
  cargo::rerun-if-changed=build.rs
  cargo::rerun-if-changed=kernels
  cargo:rerun-if-changed=/__w/rust-cuda/rust-cuda/examples/cuda/vecadd/kernels

  --- stderr
      Blocking waiting for file lock on build directory
     Compiling cuda_std v0.2.2 (/__w/rust-cuda/rust-cuda/crates/cuda_std)
  error: couldn't load codegen backend /__w/rust-cuda/rust-cuda/target/debug/deps/librustc_codegen_nvvm.so: file too short

  error: could not compile `cuda_std` (lib) due to 1 previous error

  thread 'main' panicked at examples/cuda/vecadd/build.rs:16:10:
  called `Result::unwrap()` on an `Err` value: BuildFailed
  stack backtrace:
     0: __rustc::rust_begin_unwind
               at /rustc/be19eda0dc4c22c5cf5f1b48fd163acf9bd4b0a6/library/std/src/panicking.rs:697:5
     1: core::panicking::panic_fmt
               at /rustc/be19eda0dc4c22c5cf5f1b48fd163acf9bd4b0a6/library/core/src/panicking.rs:75:14
     2: core::result::unwrap_failed
               at /rustc/be19eda0dc4c22c5cf5f1b48fd163acf9bd4b0a6/library/core/src/result.rs:1761:5
     3: build_script_build::main
     4: core::ops::function::FnOnce::call_once
  note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
warning: build failed, waiting for other jobs to finish...
error: failed to run custom build command for `sha2_crates_io v0.1.0 (/__w/rust-cuda/rust-cuda/examples/cuda/sha2_crates_io)`
note: To improve backtraces for build dependencies, set the CARGO_PROFILE_DEV_BUILD_OVERRIDE_DEBUG=true environment variable to enable debug information generation.

[LegNeato]

Yeah, that is an intermittent failure where sometimes github actions runs before the entire image or something is downloaded? Or maybe it is a race condition related to the duplicate output warning? I haven't investigated what is going on, but it usually goes away on the rebuild. It weirdly only affects the arm64 jobs.