[FEATURE] Enhance gherkin suite for custom operator edge cases and error handling consistency

[toddbaert]

Parent Issue

This is a sub-issue of open-feature/flagd#1904.

Summary

The flagd-testbed gherkin suite needs to be enhanced to cover edge cases and error handling inconsistencies identified across the in-process provider implementations. The testbed already has some error case coverage but needs significant expansion to serve as the source of truth for consistent behavior.

Background

Several consistency bugs have been identified across the SDK implementations (Go, Java, JS/TS, .NET, Python):

• [BUG] sem_ver operator parsing behavior differs across SDK implementations flagd#1873 — sem_ver parsing and comparison inconsistencies
• [BUG] Custom operator error return values are inconsistent across SDK implementations flagd#1874 — Custom operator error return value inconsistencies
• [BUG] $evaluators/$ref resolution: no parse-time validation, non-deterministic ordering, and missing Rust support flagd#1875 — $evaluators/$ref resolution gaps

Note: fractional bucketing formula differences (open-feature/flagd#1872) will be addressed separately by the high-resolution fractional bucketing work in open-feature/flagd#1903.

Required Test Additions

sem_ver operator (flagd#1873)

• v-prefixed versions: "v1.2.3" should be accepted and treated as "1.2.3"
• Partial versions: decide and test whether "1.2" / "1" are accepted (padded to X.Y.0 / X.0.0) or rejected
• Build metadata ignored in comparisons (SemVer 2.0.0 Rule 10):
  • "1.0.0+build1" == "1.0.0+build2" → true
  • "1.0.0+build" == "1.0.0" → true
• Parse failure on invalid version strings returns null

Error return standardization (flagd#1874)

All custom operators should return null on error (not false, not empty string, not exception). This ensures errors trigger a fallback to the default variant rather than taking the false branch of a conditional.

Test cases needed:

• fractional with invalid args → null (default variant)
• fractional no-bucket-match → null (default variant)
• sem_ver with unparseable version → null (default variant)
• sem_ver with unknown operator → null (default variant)
• starts_with with wrong arg types → null (default variant)
• ends_with with wrong arg types → null (default variant)

$evaluators/$ref (flagd#1875)

• Unresolved $ref → clear error / PARSE_ERROR
• Whitespace variations in $ref objects
• Nested $ref resolution

Notes

These tests will serve as the conformance suite that each provider implementation must pass. Once this suite is enhanced, each provider can vet and fix its implementation against it.