[Snapshots] PK snapshot: Persisted-cursor storage-shape decision

[kvch]

STILL NEEDS DECISION IF WE NEED IT OR NOT

What to build

Decide, in this issue, how the per-(table, sub-range) cursor is persisted. Options:

• an errors-sibling JSONB cursor column on pgstream.snapshot_requests
• a dedicated companion table keyed by (schema, table, sub-range).

Also decide the migration approach (the project's go-bindata migrations) and how cursors reconcile when the split count differs between runs.

Requirements

• Storage shape (column vs companion table) approved in this issue
• Migration approach agreed
• Split-count-change reconciliation approach noted

Depends on

• [Snapshots] PK snapshot: In-run reconnect/resume (cursor + backoff) #939 — In-run reconnect/resume (cursor + backoff)