Benchmarks; Direct access to PJRT Buffers (when using CPU); Several speed ups.

• Added install_linux_amd64_amazonlinux.sh and pre-built libraries for amazonlinux (built using old glibc support).
• Fixed installation scripts: s/sudo/$_SUDO. Also made them more verbose.
• Removed dependency on xargs in installation script for Linux.
• Improved documentation on Nvidia GPU card detection, and error message if not found.
• Updated GitHub action (go.yaml) to only change the README.md with the result of the change, if pushing to the
  main branch.
• Added prjt.arena to avoid costly allocations for CGO calls, and merged some of CGO calls for general speed-ups.
  The following functions had > 50% improvements on their fixed-cost (measured on transfers with 1 value, and minimal programs)
  execution time (not the variable part):
  • Buffer.ToHost()
  • Client.BufferFromHost()
  • LoadedExecutable.Execute()
• Added BufferToHost and BufferFromHost benchmarks.
• Added support for environment variable XLA_DEBUG_OPTIONS: if set, it is parsed as a DebugOptions proto that
  is passed to the JIT-compilation of a computation graph.
• LoadedExecutable.Execute() now waits for the end of the execution (by setting
  PJRT_LoadedExecutable_Execute_Args.device_complete_events).
  Previous behavior lead to odd behavior and was undefined (not documented).
• Package dtypes:
  • Added tests;
  • Added SizeForDimensions() to be used for dtypes that uses fractions of bytes (like 4 bits).
• Added Client.NewSharedBuffer (and the lower level client.CreateViewOfDeviceBuffer()) to create buffers with shared
  memory with the host, for faster input.
  • Added AlignedAlloc and AlignedFree required by client.CreateViewOfDeviceBuffer.
• Added Buffer.Data for direct access to a buffer's data. Undocumented in PJRT, and likely only works on CPU.
• Fixed coverage script.

v0.4.9 Optional (static) pre-linking CPU PJRT; MacOS support with static linked PJRT

• Optional preloading CPU PJRT plugin:
  • github.com/gomlx/gopjrt/pjrt/cpu/static that statically links the PJRT CPU plugin: easy to deploy binary.
    It includes the corresponding C BUILD rule to build the static library (libpjrt_c_api_cpu_static.a)
  • github.com/gomlx/gopjrt/pjrt/cpu/dynamic that dynamically links (and preloads) the PJRT CPU plugin.
• pjrt_c_api_cpu.so now compiled directly from gopjrt, and doesn't require cloning xla separately. It will
  be distributed in the same tar.gz file.
• Added MacOS support by statically linking the CPU PJRT plugin.

v0.4.8 Minor C++ code update

• Replaced C++ xla::StatusOr by absl::StatusOr (the former was already an alias to the later) -- required for upcoming XLA change.

v0.4.7 Updated XLA dependencies, PJRT v0.57

• Sync'ed with updated proto definitions from OpenXLA/XLA project.
• TestEndToEnd: added klog flags; list devices before trying to compile.
• Renamed deprecated xla::Status to absl::Status.
• Update to XLA and PJRT v0.57
  • Updated XLA dependency.
  • Updated PJRT CPU plugin.
  • Updated pjrt_c_api.h: copying over from XLA source is now part of the generate program.
  • Note: PJRT v0.56 was broken for a few days, and the version was skipped.
    (breakage here openxla/xla@590b36f#r149134910)
• Mac version broken :( : Following up on openxla/xla#19152. Since it's
  outside our control, not blocking the release here.

v0.4.6 Minor fixes and improvements

• Fix to installation script: missing sudo to remove old library, not observing the GOPJRT_NOSUDO request.
• Fixed github test action go.yaml.
• Explicitly set the random algorithm to Philox when using RngBitGenerator. Also improved documentation and added
  check on the validity of the random state shape.
• Added dtype.DType.IsUnsigned()

v0.4.5 Clean up multi-platform paths and code

• Fixes to experimental/GPU MacOS (darwin) on arm64.
• XlaBuilder works on Darwin/X86_64 (darwin_amd64) but OpenXLA/XLA PJRT CPU does not work (yet?).
• Normalized names of prebuilt-binaries.
• Test TestEndToEnd only test first device by default, because CPU PJRT seems to falsely advertise more than one addressable device.
  • Added --alldevices to loop over all devices during the test.

v0.4.4 Buffers and literals small updates

• Package pjrt:
  • Fixed some API documentation issues with Buffer transfers from host. Added tests.
• Package xlabuilder:
  • Fixed NewArrayLiteral[T dtypes.Supported](flat []T, dimensions ...int) to create a scalar if no dimensions are passed.

v0.4.3 Static XlaBuilder library; Experimental Apple/Metal support

• GoMLX XlaBuilder C library is now linked as a static library (.a instead of .so).
  • Using new Bazel 7.4.0, with support for cc_static_library.
• EXPERIMENTAL support for Apple/Metal (darwin-arm64) support:
  • Added C-wrapper compilation for darwin-arm64.
  • Added converter from HLO to StableHLO -- it greatly increases the size of libgomlx_builder.a, since it has to
    include the whole LLVM :(
    • Enables Apple Metal PJRT -- it only supports StableHLO/MLIR programs (and not the simpler HLO).
    • Only enabled for Darwin
• Updated XLA dependency; Updated PJRT for linux/amd64 CPU.
• Added Literal.Data()

v0.4.2 New operations

• Added IsFinite and PopulationCount operations.
• Updated protos.

v0.4.1 TPUs, updated XlaBuilder, improved installation

• Added memory layout information in buffer-to-host transfers: required for TPU.
• Included C error message when reporting PJRT plugin failures.
• Added GOPJRT_NOSUDO and GOPJRT_INSTALL_DIR to control cmd/install.sh and cmd/install_cuda.sh.
• Improved installation instructions to install directly from Github using curl, without the need to clone the repository.
• Updated XlaBuilder C-wrapper to refactorings withing github.com/openxla/xla.