limit size of captured health check output

[mkmik]

Hi,

I was stress testing consul with a health check script that creates hundreds of MB of output. The consul daemon started to use a lot of cpu and DNS response were timing out.

I believe it might be useful to provide a safety size limit, which should be since the notes are primarily a debugging aid.

What do you think? I'd be happy to provide a pull request of there is consensus on this issue.

[armon]

@mmikulicic Haha I can see that may cause some issues... I think probably a 1KB limit is sane. Anything beyond that is probably an abuse of the note. (I mean maybe up to like 2-4K is reasonable). We should just use a circular buffer, and capture the last LIMIT bytes.