Speed up unpacking .tar.xz files in the hermetic toolchain by using multithreaded xz execution.

Note that the multithreaded feature should be available in non-hermetic `xz` tool.

The next step will be to make `tar` and `xz` tools  available in hermetic toolchains.

Linux x86 platform operations | repository creation time, s
 :-----------: | :-----------:
download clang.tar.xz and extract with bazel       | 137
download clang.tar and extract with bazel       | 48
download clang.tar.xz and extract with non-hermetic tar (with multithreaded xz)      | 19
download clang.tar and extract with non-hermetic tar     | 38

Parameters of the Linux machine that was used to collect the data:
```
Architecture:                x86_64
  CPU op-mode(s):            32-bit, 64-bit
  Address sizes:             48 bits physical, 48 bits virtual
  Byte Order:                Little Endian
CPU(s):                      128
  On-line CPU(s) list:       0-127
```
PiperOrigin-RevId: 814354233

Author: Google-ML-Automation

.bazelrc

@@ -413,9 +413,6 @@ common:rbe_linux_x86_64_base --platforms="@ml_build_config_platform//:platform"
 common:rbe_linux_x86_64 --config=rbe_linux_x86_64_base
 common:rbe_linux_x86_64 --config=ci_linux_x86_64
 
-common:use_tar_archive_files --repo_env=USE_CUDA_TAR_ARCHIVE_FILES=1
-common:use_tar_archive_files --repo_env=USE_NVSHMEM_TAR_ARCHIVE_FILES=1
-
 common:rbe_linux_x86_64_cuda_common --config=rbe_linux_x86_64_base
 common:rbe_linux_x86_64_cuda_common --repo_env=REMOTE_GPU_TESTING=1
 # Speed up CUDA repos creation by downloading ".tar" dists from the mirror.

WORKSPACE

@@ -17,10 +17,10 @@ load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")
 # Details: https://github.com/google-ml-infra/rules_ml_toolchain
 http_archive(
     name = "rules_ml_toolchain",
-    sha256 = "bbe18d4948cd4845255ee8d400d2f29854b64d1434523052e304c1d41e90f2ec",
-    strip_prefix = "rules_ml_toolchain-3a77adf9473cf958967bc42eb9c1a663cd7ea878",
+    sha256 = "d5aef14096f367bae6f2deb5946a31892b06badd660f8c30f6565a0f9a99185f",
+    strip_prefix = "rules_ml_toolchain-18ba88facf7f0d7203adbaccb982315936b1ac7a",
     urls = [
-        "https://github.com/google-ml-infra/rules_ml_toolchain/archive/3a77adf9473cf958967bc42eb9c1a663cd7ea878.tar.gz",
+        "https://github.com/google-ml-infra/rules_ml_toolchain/archive/18ba88facf7f0d7203adbaccb982315936b1ac7a.tar.gz",
     ],
 )