LCM (support Rat and fail on lossy coercion)

[alabamenhu]

The current definition of lcm coerces arguments to Int. This generates quite unexpected behavior when passed Rat which can have a defined LCM.

Suggestion

1. Maintain current behavior if both arguments can be coerced to Int.
2. If both arguments can be coerced to Rat, do the following:

-> Rat \a, Rat \b {
  (a.numerator lcm b.numerator) / (a.denominator gcd b.denominator)
} 

3. Failing all else, fail. Lossy coercion will result in unexpected values. It is better to fail with π lcm e now (and support it later, if symbolic math were added to core) then to produce 6 (when it should be simple πe, or 8.53…). Users can always manually coerce if they want unexpected behavior.

Potential issues

• Given 1.5 and 4.5, the above algorithm will generate 4.5 (9/2). Some users may want a whole LCM, but that is more trivially generated from the rational (just the numerator) than the opposite, I think.
• Complex numbers can be calculated too, but if we fail now, support can be added later without potentially breaking code.