Make primMapListBounded into a good consumer

[clyring]

That is, good list consumer in the sense of the GHC user guide. If this can be done without a significant performance penalty when no fusion happens, I think this would be a very desirable change. My initial testing is promising, but there's more work to be done.

TODO:

• verify that this change fairly reliably allows list fusion to happen in practice
• investigate/benchmark performance implications, with and without fusion.
• update documentation
• think more carefully about and/or revert the stupid pointer arithmetic in the current draft implementation