Speed up equivalence comparisons

[viluon]

This PR improves performance on all the five benchmarks (and hopefully in real life too). It does so simply by taking some shortcuts when the left and right operands of eq, le, and lt point to the same object.

Disregard the above. I need to verify that this change has any (positive) performance impact at all. The previous benchmarks were not representative.

[SquidDev]

On master:

# Run complete. Total time: 00:23:40

PerformanceBenchmark.binarytrees  thrpt   15  3.977 ? 0.137  ops/s
PerformanceBenchmark.fannkuch     thrpt   15  5.974 ? 0.225  ops/s
PerformanceBenchmark.nbody        thrpt   15  0.599 ? 0.024  ops/s
PerformanceBenchmark.nsieve       thrpt   15  0.289 ? 0.018  ops/s

On feature/fast-equiv:

# Run complete. Total time: 00:23:15

Benchmark                          Mode  Cnt  Score   Error  Units
PerformanceBenchmark.binarytrees  thrpt   15  4.051 ? 0.180  ops/s
PerformanceBenchmark.fannkuch     thrpt   15  6.187 ? 0.309  ops/s
PerformanceBenchmark.nbody        thrpt   15  0.718 ? 0.016  ops/s
PerformanceBenchmark.nsieve       thrpt   15  0.300 ? 0.012  ops/s

This does end up being slightly faster. Couple of other changes it may be worth trying if I've a little more time:

• Reverting the changes to toDouble. Shouldn't really have any impact on performance, but we should see.
• Only applying this patch in the case of eq.