make use of scientific notation for BigFloat consistent with other types

[rfourquet]

This switches to non-scientific notation when the scientific exponent
would be in -4:5, as is the case for e.g. Float64.
For example, string(big(42.0)) is now "42.0".

This is a change I've wanted for quite a while, I don't know if I'm in the minority here. I didn't find a discussion on this topic (except for removing a useless "e+00"), so I'm not sure whether the current printing is just inherited from the choices made in MPFR, or if this was a conscious decision.
I mostly use BigFloat in a range where I find the scientific notation as an obstacle to my human parsing, e.g.

julia> big"42.0"                                                                                                         
4.2e+01

julia> rand(BigFloat, 2, 2)
2×2 Array{BigFloat,2}:
 5.59682e-01  9.24952e-01
 4.77888e-01  4.98624e-02

julia> 2 * rand(BigFloat, 3)
3-element Array{BigFloat,1}:
 1.119363387037806864843608696550048745013519509937459521981246563117370317221463
 1.557756144279027196473368290932951926003919517867431965355848779211876826219583e-01
 1.849904313908958285396788230625007253199545024377884487188520977470270959385811

In particular in the last example, it's easy for me to forget to look at the end to check if there is an exponent.
With this PR, this is turned into

julia> big"42.0"                                                                                                         
42.0

julia> rand(BigFloat, 2, 2)
2×2 Array{BigFloat,2}:
 0.559682  0.924952
 0.477888  0.0498624

julia> 2 * rand(BigFloat, 3)
3-element Array{BigFloat,1}:
 1.119363387037806864843608696550048745013519509937459521981246563117370317221463
 0.1557756144279027196473368290932951926003919517867431965355848779211876826219583
 1.849904313908958285396788230625007253199545024377884487188520977470270959385811

[JeffBezanson]

This seems pretty inefficient --- is there a way to do this via the mpfr_sprintf format string?

[rfourquet]

This seems pretty inefficient

I just did basic benchmarks to check that, and it's (only) about 35% slower.

is there a way to do this via the mpfr_sprintf format string?

I tried that but didn't find how (IIRC), so ended up with the handmade version.

[JeffBezanson]

Ok. This is a very good change, so I'll merge it.