Fix kSmallest exponent bug

[Barbarrosa]

This fixes a bug where the calculation for the cloned tree depth in kSmallest could overflow, which caused a truncation of the tree. This would start filling the resulting array with incorrect values once at least a number of results was pulled about equal to the tree depth. The remainder of the array would be filled with with undefined values.

This still appears to offer a modest performance improvement over the code in 0.7.2 while avoiding the bug that was introduced in 0.7.3. I settled on this implementation after experimenting with the following.

• Hybrid bit shifting (<<) and exponentiation (both Math.pow and **) split at ~31 bits
• BigInt conversion
• String-based bit shifting
• More conditional logic
• Looping w/ multiplication

[staltz]

Thank you!! I can confirm that this patch fixes the test suite in jitdb.