`repr(::Float128)` is broken

In Julia, `repr(x)` is supposed to return valid Julia code that can be executed to reconstruct an object equivalent to `x`. For example, `repr(::Float16)` and `repr(::Float32)` add annotations to `x` indicating its type:

```
julia> repr(Float16(1))
"Float16(1.0)"

julia> repr(Float32(1))
"1.0f0"
```

However, `repr(::Float128)` fails to do this, and evaluating its result yields a `Float64`.

```
julia> repr(Float128(1))
"1.00000000000000000000000000000000000e+00"

julia> 1.00000000000000000000000000000000000e+00 |> typeof
Float64
```

In particular, `repr(::Float128)` is completely broken for `Inf128` and `NaN128`, which are currently formatted as `inf` and `nan`.

```
julia> repr(Float128(Inf))
"inf"

julia> repr(Float128(NaN))
"nan"

julia> inf
ERROR: UndefVarError: `inf` not defined

julia> nan
ERROR: UndefVarError: `nan` not defined
```

These should format as `Inf128` and `NaN128` (see #83) respectively.