chore: build the napi binding against the published rspack_resolver crate

[stormslowly]

Why

The resolver source has moved into rspack (crates/rspack_resolver), so keeping a second copy here means every fix has to be written twice. This repository now only owns the Node.js binding, and consumes the resolver as a pinned crates.io dependency: rspack_resolver = "=0.102.0-beta.1".

Before: root package rspack_resolver (src/, tests/, benches/, fuzz/) + napi member, released as both a crate and an npm package.
After: virtual workspace with members = ["napi"]; the crate is released from rspack, the npm package from here.

What

• napi/Cargo.toml: path = ".." → "=0.102.0-beta.1"
• root Cargo.toml: virtual workspace, keeping only the release / profiling profiles
• removed the crate sources: src/, tests/, benches/, examples/resolver.rs, fuzz/
• removed the crate-only CI and release config: benchmark.yml, codecov.yml, fuzz.yml, release-plz.yml, release-plz.toml; dropped cargo publish --dry-run, the rustdoc job, the wasm32-unknown-unknown job and the cargo test steps from ci.yml
• deny.toml: allow ISC (libloading, pulled in by napi) and ignore unpublished workspace crates. The Cargo Deny job is gated on Cargo.lock changes, so it had not actually run against these two pre-existing findings before
• docs (README.md, CONTRIBUTING.md, CLAUDE.md) and justfile updated to the binding-only layout

Verified locally: cargo check --locked, cargo clippy -- -D warnings, cargo fmt --check, taplo format --check, cargo shear, cargo deny check bans licenses sources, typos, prettier --list-different, and pnpm test after pnpm run build:binding:debug (34 + 183 tests, 0 failures).