Fix precision of bigdecimal.power(integer, prec)

[tompng]

Fixes #352

Fixes this lack of precision bug

BigDecimal('0.1234').power(8, 12) #=> 0.53768e-7
BigDecimal('12.34').power(-8, 12) #=> 0.186e-8

BigDecimal('0.1234').power(8, 12)
# before: 0.53768e-7
# after:  0.537676775108e-7

BigDecimal('12.34').power(-8, 12)
# before: 0.186e-8
# after:  0.185985344038e-8

Too high precision gets lower in some case

BigDecimal('12.34').power(8.000000000001, 12)
# =>      0.537676775109e9 (doesn't change in this pull request)

BigDecimal('12.34').power(8, 12)
# before: 0.537676775108238564041e9 (precision too high)
# after:  0.537676775108e9 (precision consistent with power by non-integer)

[tompng]

Don't need to review or merge if we're doing #347 soon.