Boundary-test the decimal comparison operand grammar

[kikashy]

Goal

Add table-driven tests for the decimal-string grammar that ordered fact comparisons require.

Why

Core §2.2 fixes the operand form for greater-than, greater-than-or-equal, less-than and less-than-or-equal:

decimal = [ "-" ] ( "0" / non-zero-digit *DIGIT ) [ "." 1*DIGIT ]

That grammar refuses a leading +, a leading zero such as 007, a bare .5, a trailing 5., an exponent, whitespace, and an empty string — and it accepts -0.5 and 0.10. It also exists because JSON numbers are the wrong carrier for a business quantity whose decimal identity matters, which is a decision worth having pinned by cases rather than by prose.

The evaluator implements it. What is thin is a set of tests that walk the grammar's edges deliberately, one accepted and one rejected value per rule.

Scope

• Change test files under internal/evaluation/ only.
• Table-driven, with t.Run per case, covering at least: leading +, leading zeros, bare .5, trailing 5., exponent forms, internal or surrounding whitespace, empty string, -0, and a value with more fractional digits than the operand it is compared against.
• Assert the diagnostic each rejection produces, not merely that it was rejected.
• Do not change the grammar, the evaluator, or any schema.

Acceptance criteria

• Each rule of the §2.2 grammar has at least one accepted and one rejected case.
• Rejections assert their diagnostic code, so a rejection for the wrong reason fails.
• Scale differences that must compare equal do, and are shown to.
• env GO111MODULE=on go test ./internal/evaluation and go test ./... pass.
• Every commit includes a DCO sign-off created with git commit -s.

Contributor learning

Why a specification carries an explicit decimal grammar instead of using JSON numbers, and how to test a grammar at its edges rather than in its middle.