Bad escaping of double quote in uploaded file

Uploading a file whose name contains double quotes (") in the file, requests replaces the character with %22. This is not reversible because the % char is not escaped (it's not possible to tell apart a literal %22 from a quote), and it's different from what a browser does. Browsers seem to use backslash escape instead (tested with Firefox).

Expected Result

In a shell run:

$ nc -l 0.0.0.0 8080 | grep filename

In Python

requests.post("http://localhost:8080/", files={'file': ("""foo% 'bar" \u20ac.txt""", open("README.rst"))})

Expected, what a browser returns, in the nc shell:

Content-Disposition: form-data; name="file"; filename="foo% 'bar\" €.txt"

Actual Result

Content-Disposition: form-data; name="file"; filename="foo% 'bar%22 €.txt"

System Information

$ python -m requests.help

{
  "chardet": {
    "version": "3.0.4"
  },
  "cryptography": {
    "version": ""
  },
  "idna": {
    "version": "2.8"
  },
  "implementation": {
    "name": "CPython",
    "version": "3.8.5"
  },
  "platform": {
    "release": "5.4.0-51-generic",
    "system": "Linux"
  },
  "pyOpenSSL": {
    "openssl_version": "",
    "version": null
  },
  "requests": {
    "version": "2.24.0"
  },
  "system_ssl": {
    "version": "1010106f"
  },
  "urllib3": {
    "version": "1.25.8"
  },
  "using_pyopenssl": false
}

Reproduce the multipart upload with the provided Python request and inspect the emitted Content-Disposition header using the shown nc command. Trace the filename encoding path in Requests and its multipart dependency, then add focused coverage for quotes, literal %22, and the Unicode filename. Done means quotes use backslash escaping without making literal percent sequences ambiguous, matching the expected browser output.