feat(testbedctl): add gc-regions and purge commands

[sunng87]

Summary

Adds two more testbedctl admin commands alongside the existing flush/compact/gc, both running against the client PostgreSQL port (11043), so they work identically in distributed (haproxy), standalone, standalone-fs, and active/standby modes.

gc-regions — region-scoped GC

./testbedctl gc-regions 4398046511106                         # admin gc_regions(4398046511106)
./testbedctl gc-regions 4398046511106 4398046511117 --full     # admin gc_regions(4398046511106, 4398046511117, true)

Mirrors the table-level gc command but targets specific region IDs (u64 values, e.g. from greptimedb logs). --full/-f forces a full object-storage file listing. Region IDs must be numeric (rejected otherwise).

purge — permanently purge a dropped table

./testbedctl purge my_table        # prompts: y/N
./testbedctl purge my_table -y     # admin purge_table('my_table'), no prompt

Runs admin purge_table(...), which permanently removes a dropped table's data (irreversible). Prompts for confirmation by default; -y/--yes skips it for scripting.

Notes

• Like flush/compact/gc, these route through the frontend to the metasrv leader (check_leader). gc-regions/purge are synchronous within the GC/procedure mailbox_timeout (default 60s).
• Documented in AGENTS.md and README.md.

Verification

• bash -n testbedctl passes.
• All arg forms produce the expected SQL (verified with a stub psql): region IDs joined as numbers, --full appends the trailing boolean.
• purge confirmation flow: n aborts (exit 1), y proceeds; -y bypasses; no-arg prints usage.
• gc-regions rejects non-numeric IDs and prints usage with no args.
• testbedctl help and both docs show the new commands.