subsequence improvements

[cgay]

Andreas left a nice list of improvements needed for the subsequence code here, which I've included as a check-list below.

• Start and end are insufficiently checked. It is quite easy to generate a subsequence of negative size by subsequence(foo, start: 5, end: 3).
• Also, passing a negative start gives access to data outside the subsequence the user is allowed to see.
• The semantics of omitting the sequence end isn't well-defined, especially given stretchy source sequences.
• The good news: the above problems are not exploitable because everything is bounds-checked twice.
• The bad news: the performance leaves to be desired because everything is bounds-checked twice.
• Design decision: signal bounds error at subsequence creation time vs. element access time. The latter is more dynamic, the former gives better performance.
• Feature wish: Read-only subsequences.