Preserve native query and query-file multiplicity

[sdairs]

Summary

• preserve repeatable --query and --queries-file values in native client argv order
• reject mixed inline/file sources at clap time and guard repeated queries on pre-23.9 clients
• document the native contract and cover empty values plus -- pass-through with fake children

Tests

• cargo test -p clickhousectl --test local_client_selectors_test
• cargo test -p clickhousectl clickhouse_client
• cargo test -p clickhousectl
• cargo fmt --all --check
• cargo clippy -p clickhousectl --all-targets -- -D warnings

Closes #470

[sdairs]

Carry-over review from PR #506 (old PR for #470). This PR is the more complete implementation: it is the only one that implements issue #470's final acceptance criterion (version check so the wrapper doesn't promise an unsupported shape — the <23.9.1.1854 gate in crates/clickhousectl/src/local/mod.rs:~324-337, correctly placed after resolution for both named and direct paths, with boundary tests at 23.9.1.1853/1854). #506 punted on this with manual source inspection. Preferred over #506.

Carry-over analysis of the one substantive comment on #506: Cursor Bugbot claimed the single-flag multitoken --queries-file forwarding was wrong; sdairs refuted it with ClientBase.cpp source citations across 5 versions (multitoken(), so one flag with many paths is native-valid) while noting "Repeated flags may also parse, but one flag per file is not required." This PR adopts one-flag-per-path (mod.rs:~311-313), which is inside sdairs' validated envelope, and proves forwarding of that shape with a fake child. No adverse carry-over.

Two minor regressions vs #506 to decide on:

1. The wrapper no longer accepts multi-path-after-one-flag input (--queries-file a.sql b.sql, via num_args = 1..). The issue's failing examples used repeated flags, so acceptance criteria are still met — but confirm the narrowing is intentional.
2. Preserve native query and query-file multiplicity in local client #506 had a help-text assertion test; this PR does not. Consider adding one.

Note: the 23.9.1.1854 floor is sourced only via the linked changelog + pinned tests (plausible; repeated files predate repeatable queries). Fine, but keep the citation in the README as-is.

CI note: local postgres edge cases fails on this branch but identically on its stack base; the diff touches no Postgres code — inherited from #547's script drift, not this PR.