Multipart file support

[KeremTurgutlu]

Multipart file uploads use streaming bodies in httpx, so r.content raises RequestNotRead unless r.read() is called first to load the content into memory.

Additionally, httpx generates a random 32-character hex boundary for each multipart request, causing identical requests to produce different cache keys. We now normalize multipart content by replacing the boundary with a fixed value before hashing.

Changes:

• Call r.read() (which is no-op for byte content) before accessing r.content in _key()
• Add normalize_multipart() to replace random boundaries with a deterministic string

Example requests:

from httpx import post
from pathlib import Path

# Prepare the file
files = [("file", ('some_path', b'some text', "text/plain"))]

# Prepare form data
data = {"model": "some-model", "prompt": "transcribe this"}

# Make the request
r = post("https://example.com/upload", files=files, data=data, timeout=None)

# An example request object with files:
req = httpx.Request( "POST", "https://example.com/upload", files=files, data=data)