Reduce allocations using nil cursors and literal value cursors

[stuartcarnie]

• Replaces individual literal cursors with a single interface value. This has the benefit that the value is immediately wrapped in an iface, removing convT2I calls and subsequent allocations every time the value is read.
• Statically allocates the nil value cursors, saving further allocations as they are stateless and safe for concurrent use.

benchmark                           old ns/op     new ns/op     delta
BenchmarkIntegerIterator_Next-8     82.8          22.7          -72.58%

benchmark                           old allocs     new allocs     delta
BenchmarkIntegerIterator_Next-8     3              0              -100.00%

benchmark                           old bytes     new bytes     delta
BenchmarkIntegerIterator_Next-8     32            0             -100.00%

Required for all non-trivial PRs

• Rebased/mergable
• Tests pass

[jwilder]

@jsternberg You are more familiar with this code than I am. LGTM though.

[jsternberg]

Update the changelog and this looks good.

[stuartcarnie]

@jsternberg / @jwilder what should we put in the CHANGELOG for this one?

[jsternberg]

I'm not really sure. Likely goes in the features section and references this PR. Just a one sentence blurb about optimizing the memory usage for some cursors works. Enough for a person who is perusing the changelog to click the link for more information if they desire.