[Geneva Exporter] Eliminate schema_id collision during schema grouping

[lalitb]

Problem
We currently use a hash-derived schema_id and rely on equality of that id to decide if a schema already exists in a batch. Hashes can collide. Also, metadata.schema_ids is emitted as MD5(hex) and semicolon-separated, which isn’t what we want going forward.

What we want

1. Use a local, per-batch auto-incrementing schema_id (0,1,2,…) for each unique schema shape inside the batch.
2. Deduplicate schemas by exact schema equality, not by hash.
3. Emit metadata.schema_ids as a comma-separated list of those local ids, sorted ascending. Example: 0,1,2.
4. Keep the CentralBlob wire format the same otherwise.