XOR-based value compression for floating-point values as used in the Gorilla algorithm.

The algorithm works as follows:

1. Store the first value as-is (64 bits)
2. For subsequent values:
   • XOR the value with the previous value
   • If XOR result is 0: store single bit '0'
   • If XOR result is non-zero:
     • Count leading zeros and trailing zeros
     • If leading/trailing zero counts match the previous value's pattern: store '10' + meaningful bits only
     • Otherwise: store '11' + 5 bits for leading zeros + 6 bits for length + meaningful bits

This encoding is highly efficient for slowly changing floating-point time series data.