Update flagd-testbed to v3.6.2 and fix in-process evaluation edge cases

[toddbaert]

Summary

Update the flagd-testbed dependency to v3.6.2 and fix the in-process evaluation to pass all new edge-case tests.

Testbed v3.6.2 adds coverage for the edge cases identified in:

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

Bugs to fix

sem_ver (open-feature/flagd#1873)

1. sem_ver build metadata not ignored in comparisons; uses ==/< rich comparison operators which include build metadata, violating SemVer 2.0.0 Rule 10 (e.g. "1.0.0+build1" == "1.0.0" returns false instead of true); needs custom comparison logic that strips build metadata

Custom operator error returns (open-feature/flagd#1874)

2. Mixed None/False error returns across operators; starts_with and ends_with return None for some error types and False for others; should consistently return None

$evaluators/$ref resolution (open-feature/flagd#1875)

3. No diagnostic feedback on unresolved $refs at parse time; failure only surfaces at evaluation time via ReferenceError
4. $ref regex relies on json.dumps() output format; no tolerance for whitespace variations in input

Parent: open-feature/flagd#1904