[Jahongir-Qurbonov]

Add primitives for these bytearray operations:

• len(b): reads ob_size directly, as for bytes, list and tuple
• b[i]
• b[i] = x
• b[i:j]: covers b[i:] and b[:j] too, and always returns a bytearray

The C implementations are in bytearray_extra_ops.c.

Indexing, item assignment and slicing take the fast path only for exact bytearray objects with short int arguments. Other cases fall back to the generic implementation, which preserves:

• subclass overrides of __getitem__ and __setitem__ (the result of slicing a subclass is checked to be a bytearray, as before)
• CPython's error messages for very large indexes

Item assignment checks the value before the index, as CPython does.

On free-threaded builds, the item array is accessed while holding the object's critical section, because a bytearray can be resized concurrently.

Tests compare each operation against a generic call through Any. They cover:

• the results, the exception types and the messages
• every index around the bounds
• large positive and negative ints
• subclasses with and without overrides

They pass on Python 3.10, 3.13, 3.14 and 3.14t (free-threaded).

b[i] += x still goes through the generic path, because index assignment targets are typed as object (the existing TODO in AssignmentTargetIndex).

Micro-benchmark (2000 × 1024 operations on a 1024-byte bytearray, the same module compiled before and after this change, Python 3.13):

Operation	Before	After
len(b)	0.0051s	0.0020s
b[i]	0.0280s	0.0050s
b[i] = x	0.0319s	0.0045s
b[i:i + 8]	0.1273s	0.0454s

Part of mypyc/mypyc#644.