Do not deploy LB when average load is too small

[lifflander]

          This lower bound should not be hardcoded, but rather use either a machine-dependent absolute quantity (i.e., square root of machine epsilon) or, better yet, a lower bound that uses first principles.

Originally posted by @ppebay in #2278 (comment)

For now replace the hard-coded value with the return value of a function implemented in baseLB, make all LB algorithms use it (except for the testing ones), and have this function for now return the hard-coded value.