Float#to_s diverges from MRI: missing mantissa fraction, and the wrong fixed/exponential threshold #619

Float#to_s/#inspect (internal/object/object.go, func (f Float) ToS) formats with Go's strconv.FormatFloat(v, 'g', -1, 64) and then appends .0 only when the result contains no ., e or E. That diverges from MRI in two independent ways.

1. The mantissa loses its fraction in exponential form

Ruby always shows a fractional digit; the .0 is appended only in fixed form today, so 1e-06 keeps a bare mantissa.

2. The fixed/exponential threshold is Go's, not Ruby's

Go's %g switches to exponential by comparing the exponent against the number of significant digits. MRI's flo_to_s (numeric.c) switches on where the decimal point falls: exponential iff decpt < -3 || decpt > DBL_DIG (15).

Measured against ruby 4.0.5, the boundaries are exactly:

1e14   -> 100000000000000.0      decpt 15, fixed
1e15   -> 1.0e+15                decpt 16, exponential
1e-4   -> 0.0001                 decpt -3, fixed
1e-5   -> 1.0e-05                decpt -4, exponential
123456789012345.0   -> 123456789012345.0       decpt 15, fixed
1234567890123456.0  -> 1.234567890123456e+15   decpt 16, exponential

The visible symptom is 1e6: MRI prints 1000000.0, rbgo prints 1e+06.

Suggested shape

Take the shortest round-tripping digits from strconv.FormatFloat(v, 'e', -1, 64), derive decpt from its exponent, then choose fixed ('f', -1) or exponential by MRI's rule, and in exponential form ensure the mantissa carries at least one fractional digit. Go's exponent padding already matches (e+06, e-05, e+100, e-323).

Why it is worth a dedicated pass

The blast radius is wide — every pinned test and spec that prints a float — so it wants the full go test ./... gate and probably some pinned-value updates in the same change. It was found while landing #617 (core/process) and deliberately left out of that PR, which owns unrelated files.