Add std/dec64 and std/dec64math (Douglas Crockford DEC64 decimal floating point)

[jonasohrn]

Summary

Adds two new stdlib modules implementing Douglas Crockford's DEC64 64-bit decimal floating-point type:

• std/dec64 — type, arithmetic, comparison, hashing, $/parse, 'dec custom literal, mixed (Dec64, int) overloads, decimal-native floor/ceil/round/trunc/abs/sign/integer-exponent pow.
• std/dec64math — elementary transcendentals (sqrt, exp, ln/log, log2, log10, sin, cos, tan, arcsin, arccos, arctan, arctan2, sinh, cosh, tanh, pow(Dec64, Dec64), nthRoot, exact-integer factorial). Transcendentals delegate to std/math via a float64 round-trip, trading ~1 ulp at Dec64's last decimal digit for a pure-Nim implementation that wraps libm rather than reimplementing polynomials. A precision-preserving dec64FromFloatPrecise is exposed alongside the fast scale-and-round dec64FromFloat for callers who need shortest-round-trip fidelity.

Small footprint: ~510 lines of executable Nim across both modules, plus ~210 lines of comments. Tests are another ~350 lines.

Relation to RFC #308

This PR is a partial response to nim-lang/RFCs#308 ("Add decimal to standard library", Accepted RFC). That RFC primarily discusses IEEE 754-2008 decimal128 (128-bit, 34 digits) as the target format.
This PR instead implements Crockford's DEC64 (64-bit, ~16.8 digits) — a smaller, faster alternative aimed at money and human-facing measurements where DEC64's range and precision are sufficient.
An IEEE 754 decimal128 implementation remains a valuable follow-up and could coexist as e.g. std/decimal128; the two address different precision/size tradeoffs.

Why DEC64 in stdlib (not just Nimble)

DEC64 is a foundational numeric type — money, measurements, human-facing decimal values where binary float rounding is a bug. As a stdlib module it composes with tables, json, strutils, etc. without a Nimble dependency.

Test plan

• nim c -r tests/stdlib/tdec64.nim (orc + refc)
• nim c -r tests/stdlib/tdec64math.nim (orc + refc)
• nim c -r -d:release tests/stdlib/tdec64math.nim
• nim doc lib/pure/dec64.nim and lib/pure/dec64math.nim — runnableExamples pass
• dec64FromFloat (fast) and dec64FromFloatPrecise (string round-trip) agree to within 2e-15 relative on inexact inputs

Open to feedback on

• Dec64 vs DEC64 for the type name — currently Dec64 per Nim's PascalCase convention; the spec uses both spellings.
• Whether dec64FromFloatPrecise belongs in the public API. It duplicates dec64.nim's dec64(f: float) constructor exactly; could be removed in favor of pointing precision-conscious callers at the constructor.
• Hyperbolic functions (sinh/cosh/tanh) — included for symmetry with std/math; happy to drop if reviewers prefer a tighter surface.
• Could add Table & Set providing Dec64Table[V] and Dec64Set that canonicalize keys at the boundary, with the canonical invariant enforced by a private distinct tag so the stdlib Table/HashSet internals see only fast bit-compare hash/==.

Tests cover packing, NaN propagation, alignment overflow, the 128-bit multiplication path, division precision, parse/$ round-trip, hash equality after canonicalization, native-vs-float-trip parity, and std/math agreement to 1e-14 relative on a sampled domain.