slight change to rules for `along`

[Michael Schubert]

• For vectors, use them disregarding the value of along. This useful if we have e.g. the same ID type in a vector and columns in a matrix.
• Allow along=-1 etc. for the n-th last dimension not only in split() but throughout the package.
• Remove requirement that along has to be 1 if a data.frame field is referenced; instead, be consistent with above rule for vectors
• Allow named vector for object references

[Michael Schubert]

good enough for now