resolve: ChangeStream.ts merge conflicts for raw change streams

Re-applied all Cosmos DB changes to the new raw change stream code
structure from PR #591. The raw aggregate approach is better for
Cosmos DB: no lazy ChangeStream init, explicit cursor management,
$changeStream stage built directly in pipeline.

Changes applied to new structure:
- detectCosmosDb() calls in getSnapshotLsn, initReplication, streamChangesInternal
- getEventTimestamp() adapted to ProjectedChangeStreamDocument type
- Sentinel checkpoint with BSON.deserialize for fullDocument (raw Buffer)
- Pipeline guards: skip $changeStreamSplitLargeEvent and showExpandedEvents
- Cluster-level aggregate (admin db + allChangesForCluster) when isCosmosDb
- startAtOperationTime fix (startAfter != null)
- Keepalive guard for Cosmos DB resume tokens
- .lte() dedup guard skip on Cosmos DB
- wallTime tracking for replication lag
- Added changeset for @powersync/service-module-mongodb (minor)

Verified: 59/59 standard MongoDB tests pass.
Cosmos DB cluster is currently down — tests blocked by TLS timeout.
Code audit of RawChangeStream.ts found no compatibility issues:
cursor ID type auto-fixed by BigInt, postBatchResumeToken needs
empirical verification when cluster is back.

Author: Sleepful

.jjconflict-base-0/.changeset/README.md .changeset/README.md.jjconflict-base-0/.changeset/README.md renamed to .changeset/README.md

@@ -0,0 +1,5 @@
+---
+'@powersync/service-module-mongodb': minor
+---
+
+Add experimental Cosmos DB MongoDB vCore support. Auto-detects Cosmos DB via `hello` command and applies workarounds for missing `clusterTime`, `operationTime`, and unsupported change stream features.