Fix `rrd` not containing end marker

[jprochazk]

For some reason, the rrd file produced in test_load_recording (in test_dataframe.py) does not contain an end stream marker.

The file the test produced on my machine:
tmp.rrd.zip

pixi run rerun tmp.rrd works and contains all expected data.

I ran into this issue in

• Migrate file format to protobuf #8995

It took me a few hours to realise this was the cause of this CI failure, so I have not investigated the real reason behind it, and only applied a workaround.

The workaround was to ignore read failures at the start of a new message (ea90dd8). We've had this exact snippet in there since way before protobuf was in the picture:

rerun/crates/re_log_encoding/src/decoder.rs

Lines 122 to 130 in 2514b72

	let header = match MessageHeader::decode(&mut self.read) {
	Ok(header) => header,
	Err(err) => match err {
	DecodeError::Read(e) if e.kind() == std::io::ErrorKind::UnexpectedEof => {
	return None
	}
	other => return Some(Err(other)),
	},
	};

Note that in the past, it was only ignoring incomplete MessageHeaders, but now it ignores incomplete messages in full. This seems wrong, we should not have to do that. It may hide real errors caused by incomplete streams.