Speed up holmes by limiting data fetched from kubectl logs

[aantn]

When there are many logs, it takes a while to run kubectl logs and we throw away most of the data anyway due to token limits:

We should limit the data fetched upfront so that holmes will be faster