Add support for Parquet Input Stream optimisations.

This issue tracks adding of a new module [parquet-io] with the goal of adding IO optimisations to the parquet-java repository.

The goal is to have a single place where the following optimisations can be implemented:

* Vectored Reads
* Reading the tail of a file in one request rather than multiple small requests (Avoid the parquet footer dance, multiple requests for the pageIndex)
* Small Parquet files are read in a single request
* Sequential prefetching

[Parquet Java Input Stream Optimisations](https://docs.google.com/document/d/1Xdlh23tmCs-KvzHhY2RuwFYmc3xntUKcmwb8yxEl78Y/edit?usp=sharing): Doc explains the features this will implement.

[Analytics Accelerator for S3](https://docs.google.com/document/d/13shy0RWotwfWC_qQksb95PXdi-vSUCKQyDzjoExQEN0/edit?tab=t.0#heading=h.3lc3p7s26rnw): Doc explains IO optimisations made in the analytics accelerator library.

Start by reading the Parquet Java Input Stream Optimisations document and the Analytics Accelerator for S3 document linked in the issue. Then define the scope of the proposed parquet-io module around vectored reads, tail and small-file reads, page-index access, and sequential prefetching. Done would require an agreed implementation plan and support for the listed optimisations.

Written by the indexing model from the issue text.