Use incremental decoder for log output

[whym]

Long log lines can be split internally when it is very long. The default decoder (str.decode) raises UnicodeDecodeError against partial fragments of multi-byte characters. Incremental decoder doesn't have the same problem.

Fixes #1308

About tests:

I have done manual testing with the yaml file below:

services:
  s1:
    image: docker.io/python:3.11-slim
    command: env LANG=ja_JP.utf8 python -c'print("あ" * 10000000 + " end")'

I realize I'm expected to add unit tests, but I don't know what test to add here. Any advice?

• I think it should interact with the decoder's internal buffer size for reproducibility. I don't see that part exposed in the decoder API.
• I think it should handle potential freezes (maybe with a hard time limit of N seconds), because the potential failure mode is not wrong result, but freeze here. I don't know how to imprement that in a unit test in a clean way.