pgbench: Add --init-batch-size with automatic retry on disconnect

[dcw-data]

Add a new --init-batch-size option to pgbench initialization mode that enables batched commits and automatic reconnection/retry on errors.

The --init-batch-size parameter accepts a value in scale units (same as --scale). When specified, the initialization process commits after each batch instead of using one large transaction. This allows pgbench to recover from network disconnections or other transient errors during long-running initializations that can take hours for multi-TB datasets.

Key features:

• Batched commits: Each batch commits independently, preserving work
• Automatic retry: On error, pgbench reconnects and retries the current batch up to 5 times before giving up
• Progress tracking: Allows dozens of retries over the entire run as long as progress is made between retries
• Error reporting: Reports total errors and retries at completion
• COPY FREEZE disabled: When batching is enabled, COPY FREEZE is automatically disabled since it requires a transaction that created the table

The implementation:

• Modifies initPopulateTable() to accept a connection pointer and handle batching/retry logic
• Updates initGenerateDataClientSide() to support both batched and non-batched modes
• Maintains cancel handler (SetCancelConn/ResetCancelConn) across reconnections
• Adds global counters for num_init_errors and num_init_retries

Also adds TAP tests to verify the retry behavior works correctly when connections are terminated during initialization.

Example usage:
pgbench -i --scale=1000 --init-batch-size=10 dbname

This commits every 10 scale units (1 million accounts rows per batch).