feat(server): allow a timestamp override on identify

[santimillang]

💡 Motivation and Context

Closes #303.

capture() in posthog-server already accepts a timestamp override (#297), but identify() didn't, so backfilling or replaying identify calls meant hand-building a $identify event via capture.

This adds an overload:

posthog.identify(distinctId, userProperties, userPropertiesSetOnce, timestamp)

• PostHogInterface (sealed) gets the new 4-arg abstract identify; the existing 3-arg one becomes a default that delegates with timestamp = null, so current callers are unaffected (binary and source).
• PostHogStateless.identify now routes through a protected identifyStateless(..., timestamp) that forwards the timestamp to captureStateless. Android behavior is unchanged.
• The API surface change is additive only (apiDump regenerated).

sdk-specs: this matches the identify spec (openspec/specs/identify/spec.md), which lists options.timestamp. Go and Ruby already expose it. The only divergence is shape: the spec sketches an options object, while this follows the existing capture(..., timestamp) overload style in this SDK. Happy to switch to a PostHogIdentifyOptions builder if you'd prefer.

Note: I'm an external contributor. There's no separate agreement issue beyond #303, but the published spec covers this API.

💚 How did you test it?

• New test in PostHogTest: identify with a timestamp sends a $identify event carrying that timestamp and the $set properties (written first, failed, then passed).
• ./gradlew :posthog:test :posthog-server:test: all pass except PostHogServerWireSnapshotTest, which fails identically on main when run on Windows (it compares <project>/... paths against D:\...), so it's unrelated.

📝 Checklist

• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran pnpm changeset to generate a changeset file

🤖 Generated with Claude Code