Reusable file reader can retain stale bytes after a read exception

`input_all_with_reused_buffer` deliberately keeps a `Buffer.t` across calls to reduce allocation pressure in `get_all_procs`. The buffer is currently cleared only after a complete successful read.

If `Unix.read` raises after one or more chunks were appended, the exception leaves those bytes in the reusable buffer. The next `Blocking_IO.string_of_file` call starts appending to the same buffer and can return stale bytes from the failed file prefixed to unrelated `/proc` contents. A transient read error can therefore poison later reads in the same process.

The buffer should be empty at the start of each read and cleared in a `finally` block on both success and failure. I have a focused patch ready that makes that lifecycle explicit without changing the reuse optimization.

Start at input_all_with_reused_buffer and follow its use from get_all_procs through Blocking_IO.string_of_file. Inspect the reusable Buffer.t lifecycle around successful and exceptional Unix.read calls. Done means a failed read cannot affect the next file read while the existing buffer-reuse behavior remains intact.

Written by the indexing model from the issue text.