Revert "Replace in-house rand with absl::Random for backoff"

[jtattermusch]

Reverts #26463

Looks like the PR is breaking ARM64 tests for multiple languages:
python
https://source.cloud.google.com/results/invocations/23a0d071-ee3d-4ce1-b0cb-f65a032bfb8a/log

php
https://fusion.corp.google.com/projectanalysis/summary/KOKORO/prod:grpc%2Fcore%2Fmaster%2Flinux%2Faws%2Fgrpc_aws_basictests_php

ruby
https://fusion.corp.google.com/projectanalysis/summary/KOKORO/prod:grpc%2Fcore%2Fmaster%2Flinux%2Faws%2Fgrpc_aws_basictests_ruby

[veblush]

abseil/abseil-cpp#662 looks relevant to this

[veblush]

Jan, do you have a different fusion dashboard for ARM builds than https://fusion.corp.google.com/label/1923284992 ?

[veblush]

It seems like we need to add absl::random specific configuration to makefile.

Except from abseil-cmake:

ABSL_RANDOM_HWAES_ARM32_FLAGS = [
    "-mfpu=neon",
]

ABSL_RANDOM_HWAES_ARM64_FLAGS = [
    "-march=armv8-a+crypto",
]


if(CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|amd64|AMD64")
  if (MSVC)
    set(ABSL_RANDOM_RANDEN_COPTS "${ABSL_RANDOM_HWAES_MSVC_X64_FLAGS}")
  else()
    set(ABSL_RANDOM_RANDEN_COPTS "${ABSL_RANDOM_HWAES_X64_FLAGS}")
  endif()
elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "arm.*|aarch64")
  if (CMAKE_SIZEOF_VOID_P STREQUAL "8")
    set(ABSL_RANDOM_RANDEN_COPTS "${ABSL_RANDOM_HWAES_ARM64_FLAGS}")
  elseif(CMAKE_SIZEOF_VOID_P STREQUAL "4")
    set(ABSL_RANDOM_RANDEN_COPTS "${ABSL_RANDOM_HWAES_ARM32_FLAGS}")
  else()
    message(WARNING "Value of CMAKE_SIZEOF_VOID_P (${CMAKE_SIZEOF_VOID_P}) is not supported.")
  endif()
else()
  message(WARNING "Value of CMAKE_SYSTEM_PROCESSOR (${CMAKE_SYSTEM_PROCESSOR}) is unknown and cannot be used to set ABSL_RANDOM_RANDEN_COPTS")
  set(ABSL_RANDOM_RANDEN_COPTS "")
endif()

[veblush]

Instead of rollback, we might handle this by #26479

[jtattermusch]

Jan, do you have a different fusion dashboard for ARM builds than https://fusion.corp.google.com/label/1923284992 ?

The ARM builds weren't in the main dashboard before, but I just added them.
The impacted tests seem to have been fixed by the revert. Thanks!