add `swamp doctor extensions` subcommand for on-demand extension load diagnostics

Problem

After swamp-club#177 lands a stderr-based warning emitter for extension load failures (so warnings appear on every command that calls ensureLoaded()), users still benefit from a way to ask "is anything wrong with my extensions?" without running another command and parsing its stderr. Today the only signal is the side-effect on commands that touch the registry; there is no dedicated diagnostic.

Proposed solution

Add swamp doctor extensions as a sibling subcommand to the existing swamp doctor audit (registered in src/cli/commands/doctor.ts). The command:

• Calls ensureLoaded() across all five user registries (model, vault, driver, datastore, report) so every loader runs and any failures are recorded by the issue-177 warning collector.
• Dumps the collector contents through a renderer that supports both log and JSON output modes, matching the project rule that every command supports both modes.
• Exits non-zero when failures are present, so this fits naturally into swamp doctor aggregations and CI preflight checks.

Scope

Affected components:

• src/cli/commands/doctor.ts — register the new subcommand.
• src/cli/commands/doctor_extensions.ts — new command file mirroring doctor_audit.ts.
• src/libswamp/doctor/extensions.ts — application service that drives ensureLoaded across registries and reads the collector.
• src/presentation/renderers/doctor_extensions.ts — log + JSON renderer.

Implementation is small (~30 lines of orchestration plus the renderer) once the issue-177 collector exists.

Alternatives

• Rely solely on stderr warnings from other commands (rejected: no on-demand check).
• Surface warnings as a JSON field on every command's stdout (rejected in #177's plan: pollutes stdout contract; stderr is the right channel).

Dependencies

Blocks on swamp-club#177 — the collector mechanism must exist before this subcommand has anything to read.