fix: ensure post() completes before shareHistoricalKeys()

[axel-krapotke]

Problem

When sharing an E2EE layer, shareHistoricalKeys() found 0 Megolm session keys to share — even though content had just been posted. This meant new members joining the layer had no keys and couldn't decrypt content until the safety-net mechanism fired (requiring the sharing member to be online).

Root Cause

post() did not await its internal __post() call:

Project.prototype.post = async function (layerId, operations) {
  this.__post(layerId, operations, ODINv2_MESSAGE_TYPE)  // ← missing await!
}

So ODIN's share handler sequence:

await replicatedProject.post(id, operations)        // returned immediately!
await replicatedProject.shareHistoricalKeys(id)      // ran before posts were enqueued

The shareHistoricalKeys callback was enqueued in the command queue before the content chunks, ran first, and found no Megolm sessions yet.

Fix

• post() now awaits __post()
• __post() uses for..of instead of forEach to serialize chunk scheduling
• shareHistoricalKeys() is now async, awaits schedule()
• schedule() is now async, awaits enqueue() to guarantee FIFO insertion order

Testing

• 68 unit tests passing
• Manual testing: B shares layer → keys shared immediately → A joins (B offline) → content decrypted instantly ✅