Fix large decimal price formatting

[samsamtrum]

Fixes precision loss when formatting very large decimal strings.

StringMath.log10Floor() currently converts the integer portion through Number(...) to decide whether the value is >= 1. For very large decimal strings this loses arbitrary-precision behavior and can also overflow to Infinity, while the rest of the formatter is string-based.

This derives the magnitude directly from string digits instead, and adds a regression test for a large decimal price.

I could not run the Deno test locally because deno is not installed in my environment.

[nktkas]

Thanks for the PR, but this doesn't change anything in practice:

Number(...) on the integer part only decides whether the value is ≥ 1 — the magnitude comes from the digit count (trimmedInt.length - 1), so its overflow to Infinity never changes the result.
The only inputs where the result actually differs are below ~1e-324, where Number(...) underflows to 0. formatPrice truncates to ≤ 6/8 decimals before log10Floor runs, so it never sees them.

Closing — no reachable behavior change.