Fast non-validating reader mode to count records in Avro OCF files

[mzabaluev]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.
A query like SELECT COUNT(*) ... on an Avro data source needs no data fields, only the number of rows in the partitioned data set.
With the Avro OCF format, this information can be obtained by decoding just the block frames, presuming that the data encoding is well-formed and the number of encoded records in each block matches the one stated in the block header.

Describe the solution you'd like
Add an option method to the reader builders that would make the reader bypass decompression and decoding of Avro record data. Instead, the decoder should only parse the OCF data blocks to sum the row counts, and produce record batches with no columns, but with the row counts and metadata corresponding to the file content. This method should not be used together with with_reader_schema.

The name of the method should give sufficient warning, e.g. count_without_validation.

Describe alternatives you've considered
This behavior could be enabled when the reader schema has no fields. However, since this could lead to invalid encoded data being accepted based on the block framing, it's preferable that an explicit option is used.

Additional context
#9608 concerns the behavior when the reader schema has no fields, but validation of Avro data is performed.

[mzabaluev]

A further optimization could be enabled with reading backends that can skip. The async reader has this capability and in theory, should permit implementations that reuse the HTTP connection to the data store service to submit multiple ranged GET requests. For the sync reader, perhaps a separate "counting reader" type should be provided that would require a Read + Seek I/O parameter type.

[mzabaluev]

separate "counting reader" type should be provided that would require a Read + Seek I/O parameter type.

A better idea: a "reader with extra features" (which could be the same generic Reader sporting the extra API features when extra bounds are met on its parameter), which would also support ranged reads and HeaderInfo (#9548).