release: 1.0-rc — freeze public API, semver discipline

[ch4r10t33r]

Problem

We went from `0.1.8` to `0.1.15` in a single working session. Each one was a correct release-please cut, but the cadence reads as "still moving fast" to external consumers, and there is no public statement of which surfaces are stable vs. likely to break in the next minor.

Before any zig-libp2p announcement to the broader Zig / libp2p audience we need a 1.0-RC that:

1. Freezes the `root.zig` public re-exports as the stable surface.
2. Locks the QUIC + libp2p TLS wire (covered by tests + the cross-impl matrix, but make the commitment explicit).
3. Documents what's deliberately not stable (Noise XX SSP, async swarm, anything still under Consider std.Io async / co-scheduled swarm (zig-async-style) as follow-on to threaded runtime #57 or feat: Resource Manager (rcmgr) scope-based limits #169).
4. Adopts a written semver-breakage rule ("breaking changes only with a major") going forward.

Scope

• `docs/STABILITY.md` listing every `pub const` in `root.zig` with a stability tier (stable / preview / internal).
• `README.md` Status block updated to reference the stability doc.
• Tag `v1.0.0-rc.1` once the spec gaps tracked in feat: NAT traversal — Circuit Relay v2 + DCUtR hole punching #91 feat: AutoNAT (NAT detection) #92 feat: Kademlia DHT (kad-dht) #93 feat: WebSocket / Secure WebSocket / WebRTC transports #94 feat: Resource Manager (rcmgr) scope-based limits #169 security: third-party audit + SECURITY.md disclosure policy #170 feat: PSK / pnet (private networks) #171 are either implemented or explicitly documented as out of scope for 1.0.

Out of scope

Actual ecosystem-feature parity with go-libp2p (DHT, relay, NAT). 1.0 here means "the lean-consensus / Eth2 subset is stable", not "feature parity with the largest libp2p in the ecosystem".

Acceptance

• `docs/STABILITY.md` merged.
• `v1.0.0-rc.1` published via release-please.
• README announces the freeze; CHANGELOG documents the semver rule.