[FEA] Minimal support for LEGACY timeParserPolicy

[andygrove]

Is your feature request related to a problem? Please describe.
With spark.sql.legacy.timeParserPolicy set to LEGACY we currently fall back to CPU. This issue is for the initial work of supporting the legacy policy for format strings that are already supported in CORRECTED mode and which would work identically in LEGACY mode.

We will not support the hybrid Julian+Gregorian calendar that Spark supports in legacy mode so we need to document that and perhaps have a feature flag for enabling legacy support, since we are not fully compatible.

Describe the solution you'd like

• With LEGACY mode enabled, and with supported formats, we should run on the GPU.
• With LEGACY mode enabled, and with unsupported formats, we should fall back to CPU.

Describe alternatives you've considered
N/A

Additional context
N/A