Rewrite BigDecimal#sqrt in ruby with improved Newton's method

[tompng]

Implement sqrt in ruby with faster Newton's method implementation.

Related to #323 (doing the same rewrite with Integer.sqrt)

Example calculating BigDecimal(2).sqrt in 116 precision

x = BigDecimal(2)
y = BigDecimal(Math.sqrt(x.to_f))
y = (y+x.div(y, 18)).div(2, 18)
y = (y+x.div(y, 32)).div(2, 32)
y = (y+x.div(y, 60)).div(2, 60)
y = (y+x.div(y, 116)).div(2, 116)

Minimum loop count, minimum precision in each step.
The original c implementation was slow for large precision because it was not doing this.

Small precision: gets slower
Large precision: faster

# SQRT2
two = BigDecimal(2);

10000.times { two.sqrt(10) }
# processing time: 0.007694s → 0.073325s

10000.times { two.sqrt(32) }
# processing time: 0.020750s → 0.091189s

10000.times { two.sqrt(100) }
# processing time: 0.049758s → 0.113781s

10000.times { two.sqrt(1000) }
# processing time: 2.865216s → 0.777328s

two.sqrt(10000)
# processing time: 0.995833s → 0.017588s

# Fast path
BigDecimal(121).sqrt(10000)
# processing time: 0.000140s → 0.000154s

# Many digits
BigDecimal(2).div(7, 10000).sqrt(10000)
# processing time: 0.980485s → 0.014949s
# note: using Integer.sqrt is faster(0.002656s) in this case

[mrkn]

@tompng I like this approach. Please check a comment I left.

[mrkn]

@tompng, how do you think about extracting the infinity check and exception-raising part as a method?

[tompng]

This is sqrt-specific condition. For example, Math.atan(infinity)=π/2, `"Computation results in 'Infinity'" is not a correct message.

We can re-consider adding a common infinity check method later, for unbounded monotonic increasing function and other types, but I think it's too early for now.

[tompng]

Sorry I misread the comment. Changed to:

return BigMath._infinity_computation_result if infinite? == 1

This method is added in #389 and used in BigDecimal#power, BigMath.exp and BigMath.log.

[mrkn]

@tompng, I think this square root calculation can be faster if we use decimal shift operations here and on line 37, as we previously discussed. What about this?

[tompng]

I think it's good to use decimal shift operator. I've updated #324 as ready for review.