Avoid repeated remote schema fetches after failures

[iqdoctor]

Fixes #858.

Summary

• avoid repeated negative-cache retries only for remote http(s) schema fetch failures
• coalesce concurrent remote fetches for the same schema URL
• preserve immediate retries for local file:// schema failures
• publish the leader result to waiters and clean up in-flight ownership on cancellation
• add regression tests for remote sequential failures, remote concurrent coalescing, local file retries, and cancellation recovery

Verification

• cargo test -p taplo-common --features schema,reqwest,rustls-tls
• cargo test -p taplo-lsp --features rustls-tls --no-run