Clarify local server lifecycle help

[sdairs]

Summary

• document exact-current-directory lifecycle scope and current stop/remove omission rules
• add the canonical named workflow, generated-name retention, and project cleanup guidance
• contrast installed-version removal with named server-data removal in CLI help and README
• pin rendered lifecycle help and README removal guidance with exact tests

Tests

• cargo test -p clickhousectl local::cli::tests::server_
• cargo test -p clickhousectl local::cli::tests::readme_contrasts_version_and_server_removal
• cargo test -p clickhousectl
• cargo fmt --all --check
• cargo clippy -p clickhousectl --all-targets -- -D warnings

Closes #478

[sdairs]

Carry-over review from PR #520 (old PR for #478). This PR is substantively stronger than #520: stop-selection wording fixed to match the actual code (mod.rs:~896-927), scope sentence consistently worded on every subcommand, removal contrast present in CLI help as well as README, and exact-block assert_eq! pinning instead of #520's substring contains(). All six acceptance criteria met. But it re-adds content sdairs explicitly trimmed on #520. Required:

1. Reconcile with sdairs' two trim comments on Clarify local server lifecycle help #520 (both re-violated here). On Clarify local server lifecycle help #520, sdairs left two pending review comments, then self-addressed them in commit c764b48 ("Trim local lifecycle help"):

   • Comment 1 (cli.rs): "no need to explain obvious commands that adds nothing over the command descriptor". The trim removed the top-level sentence restating stop-all's descriptor. This PR re-adds that class of content in three places: cli.rs:~313-314 ("For project cleanup, use clickhousectl local server stop-all."), cli.rs:~450 ("Use clickhousectl local server stop-all when the intent is project cleanup."), and README.md:~293. Mitigation: issue Clarify local server lifecycle help: CWD scope, omitted-name behavior, and agent workflow #478 criterion 4 explicitly requires "Help recommends server stop-all when the intent is project cleanup", so the guidance is mandated. A reviewer applying comment 1's standard could still flag the wording as restating the descriptor — either trim to the minimal mandated form or state in the PR description that the earlier trim is being consciously relaxed to satisfy the acceptance criterion.
   • Comment 2 (README.md): "this is a readme, not agent help. this does not need to be added here". The trim removed the generated-name operational reminder from the README. This PR re-adds it nearly verbatim at README.md:~300: "Retain the returned generated name for later stop and remove commands, or pass a name positionally for a stable identity." The CLI help already carries this (cli.rs:~347). Delete the README retain-clause (or justify the reversal explicitly).

2. Note: the scope sentence is now duplicated verbatim 6 times (group + start/list/stop/stop-all/remove). Any wording change breaks 6 places — acceptable for agent-facing help, but be aware.

3. CI note: local postgres edge cases fails here; docs-only diff, so this is inherited from the stack base (Validate local Postgres start options #547 script drift). Confirm resolved after that lands and the stack rebases.

Supersede/close #520 when this merges to avoid duplicate "Closes #478".