Revert "Hide all gtest symbols in cudftestutil (#16546)"

[robertmaynard]

Description

This reverts commit ac42bc8.

We need to revert #16546 as it broke the gtest builds for cudf. Therefore gtests that actually fail wouldn't properly report an error but silently continue and report as passed.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[vyasr]

/merge

[robertmaynard]

Summary of the problems that packaging cudftestutil has on libcudf:

• If we use a shared GTest it enforces our conda package to a pinned gtest causing nightmares for all our conda users
• We can't use a shared GTest + private symbols since that is impossible ( we can't call the symbols )
• If we use a static GTest + public symbols we are bound to the GTest version via symbol ABI, causing runtime nightmares for the subset of conda users that leverage cudftestutil
• If we use a static GTest + private symbols none of the cudftestutil 'checkers' actually work ( e.g. why this revert PR exists )

[kkraus14]

• If we use a shared GTest it enforces our conda package to a pinned gtest causing nightmares for all our conda users

Outside of conda recipe / packaging semantics which can be loosened, why does this enforce a pinned gtest version?

[robertmaynard]

• If we use a shared GTest it enforces our conda package to a pinned gtest causing nightmares for all our conda users

Outside of conda recipe / packaging semantics which can be loosened, why does this enforce a pinned gtest version?

To be clear I mean pinned to the same major.minor value. The GTest releases offer no ABI stability between minor releases. So if we build against GTest 1.10 and the users has 1.15 we should expect undefined behavior to occur.

[vyasr]

Proposal to convert cudftestutil into a header-only library is in #16658