InferenceFS Plan
================

* Start from the passthroughfs implementation
* Alter the implementation so that it retrieves the contents of a file by making a Claude query that asks: "What is the most likely contents of a file with the name <filename>" and then retaining this content as long as the file is open. Ensure the query response is only the file contents and nothing else.
* When a file is closed, discard the contents
* Ignore all writes to files, as reads will always use the query results
* File length will not be known because the size of the query response is not known ahead of time. Find a solution for how to handle seeks in files when the size is not known at the time the file is opened
* The user must provide a Claude API token to use the filesystem. Implement a way for the user to pass the token.
