Dynamic scaling of core data structures

[ctb]

The default NodeGraph/CountGraph code requires preallocation of a fixed amount of memory, which cannot then be increased later. People don't like this - it's a UX issue that causes a lot of confusion.

As of 2.0, we have both the --max-memory-usage and --unique-kmer command line options in all scripts, which means at least its down to a single parameter setting that makes sense to intermediate users.

Long term, there are a variety of approaches we could take for enabling a dynamically growing data structure (sparsehash, #689; or forgetful bloom filters, #1198). This is a meta-issue for discussing possible approaches and tracking progress in general.