Implement pure rollout condition evaluator

For semantic version comparison, avoid string lexicographic comparison — split on '.' and compare integer tuples. Consider using the `pub_semver` package if already in the project, otherwise implement a 3-tuple comparison directly (it is a 10-line function). For the rollout hash, compute `crc32(utf8.encode(organizationId)) % 100` and check if the result is less than `rolloutPercent`. Dart does not have a built-in CRC32, but the `crc32` package is tiny; alternatively, use a simple polynomial hash: `organizationId.codeUnits.fold(0, (h, c) => (h * 31 + c) & 0xFFFFFFFF) % 100`.

Document the chosen algorithm clearly with a comment block — changing the algorithm later would change bucket assignments for all organizations, so it must be treated as a breaking change. The `FeatureFlagConfig` parameter should be a lightweight data class holding optional `minAppVersion`, `activationDate`, and `rolloutPercent` fields.

Pure unit tests with no mocking required — all inputs are passed as parameters. Test cases: (1) version comparison: 2.10.0 > 2.9.0, 1.0.0 < 1.0.1, equal versions return true; (2) activation date: past date passes, future date fails, exact current moment passes; (3) rollout 100% always true, 0% always false; (4) rollout hash is deterministic — same orgId called 1000 times returns same result; (5) rollout hash distribution: generate 1000 random UUIDs, verify roughly 50% fall within 50% bucket (statistical, not exact); (6) evaluate() returns false if any condition fails; (7) evaluate() returns true when all conditions pass; (8) null/absent conditions treated as passing. Target 100% branch coverage — this is a pure logic module with no external dependencies to mock.