fix(type): validate DecimalType construction and drop the init setters

[CurtHagenlocher]

Fixes #17.

DecimalType.Numeric validated; the record's generated constructor and init setters did not, so both of these produced NUMERIC(10,30) with IntegerDigits = -20:

new DecimalType(10, 30)
new DecimalType(10, 2) { Scale = 30 }

That value is not inert — it feeds DecimalTypeRules.Clamp and the Width tier selection, so it propagates into the arithmetic instead of failing at construction.

What changed

Options 1 and 3 from the issue, together. The positional record declaration is replaced with explicit get-only Precision/Scale, a validating constructor, and a hand-written Deconstruct:

• the constructor rejects precision outside 1..76 and scale above precision, so Numeric is now an alias for it rather than the only validating door;
• with no init setters, the object-initializer bypass is gone entirely;
• Numeric keeps its int parameters and checks them first, so Numeric(300, 0) is still rejected rather than truncated to precision 44 by the conversion to byte.

Cost is two comparisons at construction, behind NoInlining throw helpers so the constructor stays inlinable. Nothing in the library constructs a DecimalType per element — the kernels take the types as arguments — and the only in-repo construction site is DecimalTypeRules.Clamp, once per operation.

The remaining unvalidated value is default(DecimalType), which a struct always permits; it is now documented on the type as behaving like NUMERIC(0,0).

Compatibility

Source- and binary-breaking for anyone who used the init setters (an object initializer or with) — the removed set_Precision/set_Scale are gone from the public surface. The positional constructor keeps its signature and now throws for inputs it previously accepted. Nothing in this repo used either route.

AssemblyVersion is deliberately left at 0.3.0.0 — the csproj comment says to bump it to 0.4.0.0 at the next breaking release, and that call (plus whether this ships alone or with other breaks) is yours.

Tests

Six new cases in DecimalTypeTests: constructor rejection for scale > precision, precision 0, and precision 77 (each asserting the ParamName), constructor/Numeric agreement, Deconstruct, and an exhaustive sweep over all 1..76 precisions asserting IntegerDigits >= 0 for every constructible type and that scale = precision + 1 throws. Full suite passes on net8.0, net10.0, and net472 (549 tests).