Cloud Storage System, Part 1: File Operations

Your task is to implement a simple cloud storage system. All operations that should be supported are listed below.
Solving this task consists of several levels. Subsequent levels are opened when the current level is correctly solved. You always have access to the data for the current and all previous levels.
Requirements
Your task is to implement a simple cloud storage system that maps objects (files) to their metainformation. Specifically, the storage should maintain files and information about them, including each file name and size. This system is in-memory; it does not use the real filesystem.
The levels are cumulative:
Level 1 supports adding, retrieving, and deleting files.
Level 2 displays the largest files matching a prefix.
Level 3 adds capacity-limited users and user merging.
Level 4 backs up and restores a user's files.
To move to the next level, all tests at the current level must pass.
Note
The queries never call operations that produce a collision between a file name and a directory name.
FastPrep Operation-Sequence Adapter
FastPrep calls the function once with operations. Process the rows from left to right while preserving one shared storage state. Every row starts with an uppercase operation name followed by that method's string arguments.
Return one string-array row for every input operation:
Encode a Boolean as ["true"] or ["false"].
Encode an integer as one element, such as ["10"].
Encode None as an empty row [].
Return the formatted list from GET_N_LARGEST directly; an empty list is also [].
Multipart Series
Part 1: File Operations
Part 2: Largest Files
Part 3: Users and Capacity
Part 4: Backup and Restore
Level 1: File Operations
The cloud storage system should support file manipulation.
add_file(self, name: str, size: int) -> bool adds a new file name to the storage. size is the amount of memory required in bytes. The operation fails if a file with the same name already exists. Return True if the file was added successfully or False otherwise. The adapter row is ["ADD_FILE", name, size]. Files added this way are owned by the unlimited admin user.
get_file_size(self, name: str) -> int | None returns the size of file name if it exists, or None otherwise. The adapter row is ["GET_FILE_SIZE", name].
delete_file(self, name: str) -> int | None deletes file name. Return the deleted file size when deletion succeeds, or None if the file does not exist. The adapter row is ["DELETE_FILE", name].

Function
cloudStorageLevel1(operations: String[][]) → String[][]

Examples
Example 1
operations = [["ADD_FILE","/dir1/dir2/file.txt","10"],["ADD_FILE","/dir1/dir2/file.txt","5"],["GET_FILE_SIZE","/dir1/dir2/file.txt"],["DELETE_FILE","/non-existing.file"],["DELETE_FILE","/dir1/dir2/file.txt"],["GET_FILE_SIZE","/not-existing.file"]]
return = [["true"],["false"],["10"],[],["10"],[]]
The first add succeeds, the duplicate add fails, and the existing file size is 10. Deleting a missing file returns None; deleting the existing file returns 10; the final lookup again returns None.

Constraints
Every operation row is well formed and uses an operation available at this level.
Every numeric argument is a base-10 integer string whose value and all arithmetic results fit in a signed 64-bit integer.
File names, prefixes, and user IDs are non-empty case-sensitive strings.
The input never creates a collision between a file name and a directory name.
All users share one global file-name namespace. The admin user exists initially and has unlimited capacity.
Process operations in the supplied order. The storage starts empty.

The core is a hash map from file name to size. add_file checks for the key, returns false on a duplicate, otherwise inserts. get_file_size returns the size or None. delete_file pops the key and returns the old size or None. The trap is the encoding. Booleans become ["true"] or ["false"], integers become a one-element string row, and None becomes an empty row []. Sizes arrive as strings, so parse them to integers on add and stringify on return. Don't return a bare string instead of a list. Also don't treat a duplicate add as an overwrite. Since Ramp's later levels add prefix queries, users and backups, store each file as a small record with name, size and owner, not just a raw int. Owner is admin for now. If you freeze on the dispatch or the row format during the live OA, StealthCoder is the hedge that gets you unstuck fast.

Memorize the pattern. If you can't, run StealthCoder. The proctor sees the IDE. They don't see what's behind it.