fix: expect continue size cap

[alanshaw]

A test uploads a 5 GiB + 1 part with the AWS SDK and expects an EntityTooLarge error. The sequence on the wire:

1. The SDK sends the request headers with Expect: 100-continue, which it adds for any UploadPart over 2 MiB.
2. fasthttp, which versitygw runs on, has no ExpectHandler configured, so it answers 100 Continue immediately, before any handler runs. The client starts streaming the body.
3. versitygw's auth middleware then rejects on the declared Content-Length and returns the 400. fasthttp does not drain the unread body. The leftover body bytes are parsed as the next request, which fails, and the connection is closed.
4. The client is still writing. If its read loop has already pulled the 400 off the socket, the test passes. If the connection reset arrives first, the kernel discards the unread 400 and the write fails with broken pipe. Go's HTTP transport does not recover a response after a write error, so the SDK reports a transport failure with no status code. That is the log line: StatusCode: 0 ... write: broken pipe, want EntityTooLarge.