fix(lib): Support large body uploads with repeated read callbacks

When uploading larger amounts of data, cURL will call the read
callback incrementally multiple times to receive all the expected
data.

Previously if the size of the data to upload exceeded the size of the
initial buffer provided by cURL, the write (and thus the request)
would fail.

This changes the logic to write the data in chunks of a size that are
acceptable to cURL.

src/tests.rs

@@ -114,6 +114,24 @@ fn test_basic_auth() {
     assert!(response.is_success(), "authorized request should succeed");
 }
 
+#[test]
+fn test_large_body() {
+    // By default cURL buffers seem to be 2^16 bytes in size. The test
+    // size is therefore 2^16+1.
+    const BODY_SIZE: usize = 65537;
+
+    let resp = Request::post("http://127.0.0.1:4662/post")
+        .body("application/octet-stream", &[0; BODY_SIZE])
+        .send().expect("sending request")
+        .as_json::<Value>().expect("JSON deserialisation");
+
+    // httpbin returns the uploaded data as a string in the `data`
+    // field.
+    let data = resp.body.get("data").unwrap().as_str().unwrap();
+
+    assert_eq!(BODY_SIZE, data.len(), "uploaded data length should be correct");
+}
+
 // Tests for various other features.
 
 #[test]