[FEA] Implement shared parquet footer processing for Spark-RAPIDS

[GregoryKimball]

Is your feature request related to a problem? Please describe.
As of 25.02, the parquet TableScan stack in Spark-RAPIDS parses file footers 3 times. First, the footer is parsed in C++ spark-rapids-jni (code pointer). Second, the footer is parsed in Java by parquet-mr to process predicate pushdown. Third, Spark-RAPIDS writes a custom footer and this is parsed by cuDF.

Describe the solution you'd like
Starting with the single-file read case, we can reduce the time spent in footer processing.

topic	owner	status
Refactor Spark-RAPIDS to use cudf::io::read_parquet_metadata instead of rapids::jni::deserialize_parquet_footer	Spark-RAPIDS
Add features and refactor cudf::...aggregate_reader_metadata to meet the needs of Spark-RAPIDS	cuDF
Refactor Spark-RAPIDS to use cuDF's prediction pushdown utility filter_row_groups (code pointer) instead of parquet-mr	Spark-RAPIDS
Add features and refactor filter_row_groups to accommodate Spark conventions	cuDF
Update the cuDF chunked parquet reader to accept aggregate_reader_metadata as a reader option to skip parsing the file footer	cuDF

Additional context
This issue is part of larger work to refactor TableScan in Spark-RAPIDS to improve support for high selectivity queries.