Do not round exact values up under :up rounding

[b-erdem]

Follows up on the :up bug I noted in #236 — independent of #237 and #238.

Problem

:up (round away from zero) increments the coefficient whenever the discarded part is non-empty, even when all discarded digits are zero. An exact value with trailing zeros beyond the precision is rounded up:

Context.with(%Context{precision: 1, rounding: :up}, fn ->
  Decimal.add(Decimal.new(0), Decimal.new("9.0"))   #=> Decimal.new("10")   # value is exactly 9
end)

Decimal.round(Decimal.new("9.0"), 0, :up)           #=> Decimal.new("10")   # should be 9

Per the General Decimal Arithmetic spec, round-up leaves the value unchanged when every discarded digit is zero. Python's decimal returns 9 for both. :ceiling and :floor already guard on any_nonzero?/2 — so today, for a positive value, :up and :ceiling disagree even though both round away from zero (Decimal.add(0, ~d"9.0") at precision 1 gives 10 under :up but 9 under :ceiling). They must agree.

Fix

increment?(:up, …) now checks any_nonzero?(remain, sticky?) instead of returning true unconditionally — the same check :ceiling/:floor use. One clause.

Verification

Differential fuzz against Python's decimal (the GDA reference), comparing numeric values: 0 mismatches across ~3,800 :up cases spanning add/sub/mult/div and round/3, at precisions 1–5. Genuine round-ups still round up (add(0, 3.1) at precision 1 → 4; round(9.01, 0, :up) → 10).

Tests

Extends the existing with_context/2: up test with the exact-value cases (and a nonzero-discarded case that still rounds up). Full suite passes (101 doctests, 27 properties, 101 tests).

Notes

• Independent of Carry the division remainder into rounding as a sticky bit (#236) #237 and Re-round coefficient when a rounding carry exceeds precision (#236) #238: different clause, and the three branches merge cleanly (code) in any order.
• I've intentionally left the CHANGELOG entry out of this PR to avoid a conflict with the ## Unreleased sections added by Carry the division remainder into rounding as a sticky bit (#236) #237 and Re-round coefficient when a rounding carry exceeds precision (#236) #238 — happy to add one (or fold all three into a single entry) once you've decided merge order.