chore(deps): bump jsonschema from 0.18.3 to 0.58.6

[e54-bot]

Supersedes #469 (not closed here). That branch was conflicted and did not port agent_contract_vocabulary.rs.

Change

• jsonschema 0.18 → 0.58 (resolves to 0.58.6); used only in wright-cli tests.
• JSONSchema::compile → validator_for; validate → iter_errors. No assertions removed.

Strictness comparison

• 0.18.3 builds without its draft202012 feature by default, so the $schema: .../2020-12/schema in both schemas was unrecognised and the validator fell back to Draft 7. 0.58 honours 2020-12.
• Both schemas use no 2020-12-only keywords (prefixItems, unevaluated*, dependent*, $dynamicRef, array items). Their $ref siblings are only description, so Draft 7 vs 2020-12 sibling handling does not change results.
• No format keyword is used as an assertion, so format-assertion defaults do not apply.
• Result: same effective strictness; all invalid-case assertions still pass.

Verification

cargo fmt --check, cargo clippy -p wright-cli --all-targets -- -D warnings, cargo test -p wright-cli: all pass.

[Teakowa]

Verdict: changes requested (round 1). The port is correct. One dependency-footprint issue should be fixed before merge.

Blocking: crates/wright-cli/Cargo.toml:39 takes the default features of jsonschema 0.58.
They enable HTTP/file $ref resolution and pull in reqwest, rustls-platform-verifier, aws-lc-rs/aws-lc-sys (C build, needs cmake), jni and more. That is about 262 of the 442 added Cargo.lock lines, and it is all for a dev-dependency used on local, self-contained schemas. Failure scenario: every test build on every runner now compiles aws-lc-sys and needs cmake, and a schema $ref to a remote URL would make a test fetch from the network. Fix: jsonschema = { version = "0.58", default-features = false }. I checked this on the branch: agent_contract (13), agent_contract_vocabulary (2) and diagnostics_schema (1) all pass and the lock shrinks by 262 lines. I did not commit the change.

Checked and fine

1. jsonschema appears only in wright-cli [dev-dependencies] and three test files (agent_contract.rs, agent_contract_vocabulary.rs, diagnostics_schema.rs). No user was missed, including agent_contract_vocabulary.rs.
2. Strictness: the root of wright-agent-v1.schema.json holds only $schema, $id, title, $defs and $ref. The tests overwrite the root $ref, so Draft 7 versus 2020-12 sibling handling cannot change results. This agrees with the PR body. Format assertions are not used.
3. Mutation: I added a required zzz to EditRange in the agent schema and to the check schema. agent_contract and diagnostics_schema both failed. The existing invalid-instance loop in agent_contract_vocabulary still passes.
4. The diff is a mechanical API rename (compile to validator_for, validate to iter_errors). No assertion was dropped or weakened.
5. Local cargo fmt --check, clippy -p wright-cli --all-targets -D warnings and the three test targets pass. Full cargo test -p wright-cli was not run by me.

Unconfirmed: at review time CI still had Rust (stable) / Quality, Product integration and some dist-validation jobs pending.

[Teakowa]

Approve (round 2). The blocker is fixed.

• Diff: e897d6bf changes only crates/wright-cli/Cargo.toml (default-features = false) and Cargo.lock (+3/−263).
• Dependency footprint: aws-lc-sys and cmake are gone from Cargo.lock. reqwest remains once, as on main, and cargo tree -i shows it does not come from jsonschema. The lock has no cmake entry, so the build needs no C toolchain for TLS.
• Tests: locally, agent_contract (13), agent_contract_vocabulary (2) and diagnostics_schema (1) pass. cargo fmt --check and clippy -p wright-cli --all-targets -D warnings are clean.
• Mutation: with a required zzz added to EditRange in the agent schema and to the check schema, agent_contract and diagnostics_schema both fail. Invalid instances are still rejected.

Not confirmed: PR CI (including Quality, Build dist and Product integration) was still pending at review time. I ran the three jsonschema test targets, not the full cargo test -p wright-cli.