Build tool calls through ToolId instead of name literals#4156

Why

The program goal is to untangle core/ so agents read less per change, break less without noticing, and debug faster. The tool catalog names the same tool in two ways: the tools! macro generates ToolId and ToolId::name(), while about 30 call sites in ployz-store/src/tools/ spell the same wire name as a string literal (Call::new("server_show", ..), the CLI_NEXT and NOT_SETTINGS tables, LOCAL_NEXT). Renaming or adding a tool meant finding every literal by grep, and a typo only showed up as a refusal an agent could not follow.

What changed

• Call::to(ToolId, input) builds a Call from ToolId::name(). Call::new(&str, ..) stays for the sites that carry a name the caller was given (&call.tool, &state.tool).
• Every production Call::new("<literal>", ..) in tools/ now uses Call::to: act, copies, deployment, environment, github, metrics, project, release, removal, runtime, machine.
• The name tables hold ToolId instead of &str: CLI_NEXT rows (CliNext::Tool), NOT_SETTINGS, not_setting, not_setting_sentence, Tool::LOCAL_NEXT and Cx::local_next, removal::Target::tool, the stale read in machine.rs.
• New tools/tool_id_tests.rs: ToolId::name() equals the registered Tool::NAME for every tool, in catalog order, with unique names; and every string literal that is the first argument of a Call::new in tools/ equals some ToolId::name().
• ToolId variants are not renamed, there is no ToolId::of<T>, the dashboard OPERATION_TOOLS is untouched, and tool_input! (F10b) is not started.

Bets

None touched.

Behavior

Unchanged. Tool names on the wire are byte-identical: Call::to goes through Tool::NAME, and the tests that pin the wire literals (refusal_tests, machine_tests, removal_tests, agent_tool_calls) still spell the names as strings and pass untouched, apart from the two assertions on not_setting, which now compare ToolId::name to the same literals.

Proof

Baseline on the unmodified base, then the same on the branch head:

• cargo test -p ployz-store --test agent_tool_calls --test agent_tools: 92 passed and 3 passed, both before and after.
• cargo test -p ployz-store --lib tools::: 56 passed before, 58 after (the two new tests).
• cargo clippy -p ployz-store --all-targets -- -D warnings: clean. cargo fmt --check: clean.
• Sweep of the checkout for skill, verify-model, ployz-verify and doc citations into the touched files: none cite ployz-store/src/tools/*.rs by line. dashboard/src/modules/agent/agent-secrets.ts cites tools/service.rs by symbol (secret_value), which is untouched.

Residuals

• core/crates/ployz/src/handlers/mcp.rs still builds a tools::Call::new(..) from a literal (outside this PR's file list).
• The new literal scan reads Call::new( followed by a string literal; a name built another way (a variable, a format!) is not scanned.
• tools/*_tests.rs keep their Call::new("..") literals on purpose, as independent wire pins.