Fix dump/load bigdecimal with few or large precs

[tompng]

BigDecimal._load should not trust MaxPrec part of the dumped string.

Fixes this bug.

irb(main):001> x = BigDecimal(BigDecimal._load("30:"+'1'*100), 0)
=> 0.11111111111111111111586024528000051042000000087e100
irb(main):002> x
=> 0.111111111111111111100000000500000000042949672954294967295429496729542949672950000000050000000004294967295e100
irb(main):003> BigDecimal(BigDecimal._load("9999999999:"+'1'*100), 0)
# (slow, takes about a second)

Now that BigDecimal._load ignores precision part, there is no need to expose internal MaxPrec to the dumped string. Using Prec is better.

# Before
BigDecimal(0)._dump #=> "9:0.0"
BigDecimal('0')._dump #=> "18:0.0"
some_zero._dump #=> "123456789:0.0"

# After
zero._dump #=> always "9:0.0"