Support reflink (copy-on-write) based table cloning in the backup workflow to enable near-instant database-level cloning

Background

Near-instant cloning / branching of databases is becoming a common requirement:

• Gaming: A rollback point is created before every release so that the service can be quickly rolled back when a release goes wrong. Today this relies on full logical or physical copies, whose time and storage cost both grow linearly with the data size.
• AI agent sandboxes: An agent takes a snapshot before a high-risk step, rolls back on failure, forks from a snapshot to try alternative approaches, and discards it when done. These workloads are typically single-node and require replicas to be created and destroyed within seconds.

Proposal

Add a capability to WiredTiger: within the backup workflow, clone a caller-specified set of tables via filesystem reflink and register the new tables into WiredTiger.wt.

Draft interface :

/*
 * pairs:  A list of (source uri, new uri), for example
 *         "table:collection-0-123=table:collection-9-123,table:index-1-123=table:index-10-123"
 * config: method=clone|auto
 *         clone: reflink must be used; fail immediately when it is not supported
 *                (never silently fall back to a full copy).
 *         auto:  prefer reflink, and fall back to a regular copy when it is not supported.
 */
int WT_SESSION::clone_tables(WT_SESSION *session, const char *pairs, const char *config);