Improvement: sample by attribute key, message or any hash

Some ideas:

slogsampling.ThresholdSamplingOption{
	Rate: 100,
        ByHash: func(r *slog.Record) []byte {
              return []byte(r.Message)
        },
}

slogsampling.ThresholdSamplingOption{
	Rate: 100,
        ByHash: func(r *slog.Record) []byte {
              return BloomFilter([]byte(r.Message))
        },
}

slogsampling.ThresholdSamplingOption{
	Rate: 100,
        ByAttributeKey: []string{"http", "request", "header", "cf-connecting-ip"},
}

slogsampling.ThresholdSamplingOption{
	Rate: 100,
        BySource: true,   // might be very costly to look for stacktrace
}

slogsampling.ThresholdSamplingOption{
	Rate: 100,
        ByDistance: slogsampling.LevenshteinDistance(xxx),   // just brainstorming - definitely a VERY bad idea 😁
}

We have to keep track of hash/keys. A garbage collector must remove old entries in LRU or LFU manner.

The bloom filter could be embedded into the library on behalf of the developer.

Start by reviewing ThresholdSamplingOption and the existing slog.Record sampling flow. Compare the proposed ByHash, ByAttributeKey, BySource, and ByDistance ideas, then determine the intended key-retention and LRU/LFU cleanup behavior. Done should mean an agreed sampling API and implementation scope, including whether bloom-filter support is required.