Fix: EncodeAggrChunk created corrupt block if the last chunk is empty

[daijy]

This PR fixes a decode failure for aggregate chunks when the last aggregate (AggrCounter) is missing.
In this case, decoding could return invalid size instead of ErrAggrNotExist. Decoder logic expects at least one byte (+1 for encoding marker) even for zero-length entries during pre-check. For the final nil aggregate, that extra byte does not exist in the stream.

• I added CHANGELOG entry for this change.
• Change is not relevant to the end user.

[GiedriusS]

Does this ever happen in practice or is this theoretical?

[MichelHollands]

Does this ever happen in practice or is this theoretical?

We have seen it in production. The cause is still unclear.

[daijy]

It happens when the AggrCounter is missing for the AggrChunk, as shown in the test case. In AggrChunk.Get, it erroneously return "invalid size" instead of ErrAggrNotExist. Should be quite frequent, severity depends on how the caller response. In our parquet downsample converter, it stops converting the rest of tsdb downsample. I fixed it in the write side to write extra bytes for AggrCounter if it's empty, so the reader won't trigger "invalid size".