Fix for 22326: BitOps.TrailingZeroCount has inconsistent software fallback

[grant-d]

Fixes #22326
cc @tannergooding @benaadams

• Recent code in SpanHelpers uses a intrinsic with a software fallback, but the latter returns a different answer to the former for an input of 0.
• Benchmarking showed a branchless expression was slower (1.4x) than an idiomatic guard clause, so a simple if was introduced.
• All BitOps methods pass units

[grant-d]

@tannergooding or @benaadams please advise what to do about the one failed check - it looks spurious

[benaadams]

Windows_NT x64 Checked CoreFX Tests
System.IO.Tests.FileSystemWatcherTests_netstandard17
.EndInit_ResumesPausedEnableRaisingEvents

MESSAGE:
Deleted event did not occur as expected\r\nExpected: True\r\nActual: False
+++++++++++++++++++
STACK TRACE:
at System.IO.Tests.FileSystemWatcherTest.ExecuteAndVerifyEvents(FileSystemWatcher watcher, WatcherChangeTypes expectedEvents, Action action, Boolean assertExpected, String[] expectedPaths, Int32 timeout) in D:\j\workspace\windows-TGrou---74aa877a\src\System.IO.FileSystem.Watcher\tests\Utility\FileSystemWatcherTest.cs:line 261
 at System.IO.Tests.FileSystemWatcherTest.ExpectEvent(FileSystemWatcher watcher, WatcherChangeTypes expectedEvents, Action action, Action cleanup, String[] expectedPaths, Int32 attempts, Int32 timeout) in D:\j\workspace\windows-TGrou---74aa877a\src\System.IO.FileSystem.Watcher\tests\Utility\FileSystemWatcherTest.cs:line 179
 at System.IO.Tests.FileSystemWatcherTest.ExpectEvent(FileSystemWatcher watcher, WatcherChangeTypes expectedEvents, Action action, Action cleanup) in D:\j\workspace\windows-TGrou---74aa877a\src\System.IO.FileSystem.Watcher\tests\Utility\FileSystemWatcherTest.cs:line 134
 at System.IO.Tests.FileSystemWatcherTests_netstandard17.EndInit_ResumesPausedEnableRaisingEvents() in D:\j\workspace\windows-TGrou---74aa877a\src\System.IO.FileSystem.Watcher\tests\FileSystemWatcher.cs:line 208

@dotnet-bot test Windows_NT x64 Checked CoreFX Tests

[grant-d]

@dotnet-bot test coreclr-ci please

[grant-d]

From reading other PRs, it looks like the coreclr-ci test failure is spurious.
No more changes expected from my side, can we merge

[tannergooding]

LGTM.

Might be good if @jkotas, who signed off on the original PR, also approves before this is merged.

[grant-d]

Can we merge?

[jkotas]

Thanks

[grant-d]

@tannergooding, @jkotas should we look at implement the TZCNT/REP BSF and LZCNT/31 ^ BSR optimizations suggested here:
https://github.com/dotnet/corefx/issues/32075#issuecomment-462078039 ?

[tannergooding]

AFAIK, there doesn't exist any intrinsic or recognized pattern for BSF or BSR