feat(tools,scheduler): agent time-awareness — get_current_time tool + schedule add --run-at

[bug-ops]

Summary

• Adds a chrono-free, mockable ClockSource abstraction (zeph-common), reusing the existing timestamp.rs RFC3339 formatting helpers instead of adding a chrono dependency.
• Adds a get_current_time tool (zeph-tools) built on that clock, wired into all three production executor chains (CLI, daemon, ACP) via one shared clock instance.
• Adds an opt-in periodic time-reminder injection (agent.time_reminder_enabled / agent.time_reminder_interval_requests, default off) into the volatile system-prompt block, counted in agent turn-cycles rather than literal per-model-requests (the mandated injection hook is architecturally unreachable at per-request granularity — see PR discussion).
• Adds zeph schedule add --run-at <rfc3339>, exposing the existing TaskMode::OneShot scheduler primitive that previously had no CLI or tool-level entry point.
• Fixes Scheduler::init() hydration, which previously skipped one-shot rows entirely (a CLI-added --run-at job never fired), and — once that was fixed — was found to discard the user's prompt via a Value::Null config on hydration (a kind="custom" one-shot job fired a generic fallback instead of the real prompt). Both are now fixed with regression tests.

Breaking change (pre-1.0.0)

zeph schedule add positional argument order changed from <CRON> <PROMPT> to <PROMPT> [CRON]. This was required by clap, which rejects an optional positional preceding a required one once --run-at made the cron argument optional. An old-order invocation now fails loudly on cron validation rather than silently misparsing. Documented in CHANGELOG.md.

Scope note

Spec .local/specs/070-agent-time-awareness/spec.md's US-004 (an agent-callable one-shot reminder-creation tool, as opposed to CLI-only) is deferred — the existing scheduler native tool already handles natural-language deferred/periodic scheduling and was left untouched. This PR fully delivers issue #6361's three literal asks (current-time tool, periodic reminder, one-shot CLI scheduling) and closes it.

Process note

This went through 2 rounds of architecture critique (a scheduler-hydration blocker was found and fixed before implementation started) and 2 rounds of implementation critique (a critical hydration prompt-loss defect and a significant clock-sharing gap were found post-implementation and fixed), plus clean independent security and performance audits.

Test plan

• cargo +nightly fmt --check
• cargo clippy --profile ci --workspace --all-targets --features "desktop,ide,server,chat,pdf,scheduler,testing" -- -D warnings
• cargo nextest run --config-file .github/nextest.toml --workspace --features "desktop,ide,server,chat,pdf,scheduler" --lib --bins — 14323 passed, 0 failed
• RUSTFLAGS="-D warnings" RUSTDOCFLAGS="--deny rustdoc::broken_intra_doc_links" cargo doc --no-deps --workspace --features "desktop,ide,server,chat,pdf,scheduler"
• cargo test --doc --workspace --features "desktop,ide,server,chat,pdf,scheduler"
• gitleaks protect --staged
• Testing playbook: .local/testing/playbooks/agent-time-awareness.md
• Coverage rows added to .local/testing/coverage-status.md

Closes #6361