Ensure error.bufferedData is as full as possible

[ehmicky]

When the stream is larger than maxBuffer, the last chunk is currently missing from error.bufferedData, since it would make error.bufferedData larger than maxBuffer.

This PR adds the last chunk instead, but truncated so it fits within maxBuffer.

This has several advantages:

1. This gives additional information to the error to help users debug the underlying issue.
2. This prevents error.bufferedData from being empty when a single chunk was emitted and it was larger than maxBuffer. This can happen when maxBuffer is low, or when the stream highWatermark is high.
3. This ensures error.bufferedData is always exactly the same. For example, I was trying Execa with get-stream@8 and the output of childProcess.stdout was inconsistent between test runs, since the underlying stream's chunks are timing-dependent. Therefore, the test was harder to reproduce consistently.