[ML] Disregard influencers when in hard_limit

[valeriy42]

While processing the bucket data in CBucketGatherer::addEventData, we transiently fill up the queue of m_InfluencerCounts. https://github.com/elastic/ml-cpp/blob/b2e749beffff29223c3cd7dd265b3b5f7ff37bd0/lib/model/CBucketGatherer.cc#LL300C29-L300C29 When the job is hard_limit state, this can lead to temporarily using more memory than is allowed. In order to improve the stability and adherence to the memory limit, we can avoid populating this queue in the hard_limit state.