Improve `alpha test peers`: match Charon's probes; reduce timeouts

Summary

pluto alpha test peers is architecturally different from Charon's testpeers.go: instead of imperative per-test probes it runs one passive swarm event loop and synthesizes verdicts from accumulated state afterwards (build_peer_results). Concrete defects:

• PingLoad doesn't load-test: it filters passively-observed keepalive RTTs (in-code admission at peers.rs#L944-L963); Charon spawns a new ping goroutine every second for the load window. The "Running ping load tests..." log lines wrap a phase where no pinging occurs.
• Ping is a connectivity check (any connection, even a relay circuit, passes); PingMeasure grades .last() of whatever keepalive pings happened to land — no on-demand ping exists.
• DirectConn dials advertised addresses once (one-shot bool), doesn't use the existing crates/p2p/src/force_direct.rs, never checks Charon's len(conns) >= 2 invariant, and with no DCUtR behaviour in PlutoBehaviour it can't hole-punch — behind NAT it always reports failure.
• Hangs: an unreachable peer keeps peer_is_done false, so the loop runs to the global --timeout (default 1h); afterwards keep_node_alive runs for --keep-alive (default 30m). A default-flag run against a dead ENR occupies ~90 minutes. Charon bounds each test individually.
• Self is excluded from results (Charon includes the local peer); Libp2pTCPPortOpen always fails on default flags (no --p2p-tcp-address default); scoring is a placeholder TODO.
• alpha test all panics: all.rs#L34-L45 is unimplemented!() reachable from the CLI. The whole test module sits under #![allow(dead_code)] with a "foundation" TODO.

Proposed change

• Add active probes: an on-demand ping path (libp2p ping behaviour request or a dedicated protocol), used by Ping/PingMeasure/PingLoad with Charon's cadences; rewrite DirectConn around force_direct with per-second retries and the 2-connection assertion.
• Bound each test by its own timeout so unreachable peers fail fast; decouple keep-alive from the test-phase cancellation token.
• Include the local peer; give --p2p-tcp-address a sensible default or make the self-test skip when unset.
• Implement alpha test all per Charon's runTestAll: run beacon → validator → MEV → infra → peers sequentially, forcing each sub-config's quiet flag, and write all collected results at the end when the top-level --quiet is unset. Port the two pre-run guards — must_output_to_file_on_quiet and the --test-cases rejection — and resolve the list_test_cases(TestCategory::All) TODO, which returns an empty vec. (Removing the subcommand until this lands is still better than shipping the panic.)

Acceptance

Against one reachable and one dead ENR: results print within the per-test timeouts, ping tests measure actively, and alpha test all runs all five categories with Charon's quiet semantics instead of panicking.

Start with crates/cli/src/commands/test/peers.rs, especially build_peer_results and peer_is_done, then compare Charon's testpeers.go. Read crates/p2p/src/force_direct.rs and the entry points in all.rs and test/mod.rs. Done means per-test timeouts, active ping behavior, direct-connection checks, and alpha test all completing the stated reachable/dead-ENR acceptance run without panicking.