[BUG] Minimize false positives when falling back to CPU for end of line/string anchors and newlines

[anthony-chang]

Describe the bug
The changes in #5610 add very broad checks for patterns containing end of line anchor $ immediately next to a newline, begin-of-line anchor ^ or repetition that could produce empty results. However, this results in false positives:

1. $ and ^ inside character classes are being treated as anchors, eg. the pattern 0*[D$3] has no string anchors but will still fallback to CPU.
2. We are checking the entire AST node before/after a $, meaning something like \na$ is allowed, but (\na)$ will fallback to CPU. We should instead check only the component in the group that is closest to the $.
3. We fallback to CPU for cases such as ^([a-z]*)([0-9]*)([a-z]*)$ which have no line terminator characters.

Steps/Code to reproduce bug
Test with the above examples.

Expected behavior
We should support these cases on GPU, and they should be consistent with CPU

Environment details (please complete the following information)
N/A

Additional context
None.

[andygrove]

Examples in RegularExpressionSuite that we want to support

  // https://github.com/NVIDIA/spark-rapids/issues/5659
  testGpuFallback("String regexp_extract regex 1",
    "ProjectExec", extractStrings, conf = conf,
    execsAllowedNonGpu = Seq("ProjectExec", "ShuffleExchangeExec")) {
    frame => frame.selectExpr("regexp_extract(strings, '^([a-z]*)([0-9]*)([a-z]*)$', 1)")
  }

  // https://github.com/NVIDIA/spark-rapids/issues/5659
  testGpuFallback("String regexp_extract regex 2",
    "ProjectExec", extractStrings, conf = conf,
    execsAllowedNonGpu = Seq("ProjectExec", "ShuffleExchangeExec")) {
    frame => frame.selectExpr("regexp_extract(strings, '^([a-z]*)([0-9]*)([a-z]*)$', 2)")
  }