Add reminder run command

[Aaronontheweb]

Summary

Adds netclaw reminder run <id>. This command lets an operator run an
existing reminder now, so they can test it without a wait for its
schedule. It waits for the run to finish and prints the result.

• POST /api/reminders/{id}/run uses the same Operator-only authority
  gate as reminder create (ResolveReminderAuthorizationContext /
  PrincipalClassification.Operator). A non-Operator caller gets a
  403.
• The daemon runs the reminder's real prompt, to its real delivery
  target, through the same execution path a scheduled run uses.
• The endpoint waits for the run to settle, then returns: the
  reminder id, source (manual), status (ok / failed /
  timed_out), start time, duration, session id, delivery target (or
  none), the session's final assistant reply text (trimmed to a
  sane length, with a flag when it trimmed), and the error message on
  failure.
• The wait has a bound: the execution actor's own attempt timeout
  (ReminderExecutionActor.ExecutionAttemptTimeout, one hour) plus a
  settlement margin. If the run has not settled by then, the endpoint
  returns a timed_out result with the session id. It does not
  cancel the run.
• The CLI prints a short result block (status, duration, session,
  delivery, reply or error) and the exact netclaw chat --resume <id>
  line to open the session. Exit code is 0 for ok, non-zero for
  failed or timed_out. --json prints the result object. The
  CLI's HTTP client timeout for this call exceeds the daemon's own
  wait.
• Fixed a bug this work exposed: ReminderExecutionActor subscribed
  only to OutputFilter.TextStreaming. A provider that returns its
  whole reply in one chunk, with no incremental deltas, never filled
  the reply text. It now subscribes to both Text and
  TextStreaming; ExecutionOutputAccumulator's existing guard
  against double counting keeps a provider that streams both correct.
• A manual run does not change the schedule. A one-shot reminder keeps
  its original fire time. An interval or cron reminder keeps its next
  fire time.
• History records the source of each run, manual or scheduled.
  netclaw reminder history <id> and netclaw reminder status <id>
  show it in a new column.
• A manual run failure does not count toward the scheduled
  consecutive-failure limit (auto-disable).
• If the reminder already runs, the manual run fails with a clear
  error. It does not queue, and it does not reintroduce the global
  concurrency cap that fix(reminders): remove execution cap so reminders never skip on capacity #1839 removed.
• Updated the netclaw-operations skill's scheduling reference: the
  Manual run section now covers the result block, how to read each
  field, the exit codes, and the timeout case. Added a subsection on
  testing a reminder inside a session with set_reminder and
  delivery_kind=current_session. Bumped the skill version twice on
  this branch (2.75.3 -> 2.76.0 -> 2.77.0).

Design choice to check

A manual run's settlement (SettleManualExecution in
ReminderManagerActor) logs success or failure but does not post a
channel failure notice the way a scheduled failure does. The operator
now gets the full result directly from the CLI, so a duplicate channel
post felt like noise for a run the operator asked for directly. Flag
if you want manual failures to also post to the reminder's channel.

Validation

• dotnet build Netclaw.slnx -c Release: 0 errors, 0 new warnings
  (one pre-existing ASPIRE010 warning, unrelated to this change).
• dotnet test src/Netclaw.Actors.Tests -c Release --filter "FullyQualifiedName~Reminder":
  210 passed, 0 failed.
• dotnet test src/Netclaw.Daemon.Tests -c Release --filter "FullyQualifiedName~Reminder":
  20 passed, 0 failed.
• dotnet test src/Netclaw.Cli.Tests -c Release --filter "FullyQualifiedName~Reminder":
  6 passed, 0 failed.
• dotnet slopwatch analyze: 0 issues.
• pwsh ./scripts/Add-FileHeaders.ps1 -Verify: all files have headers.
• Manual end-to-end proof: isolated NETCLAW_HOME, HOME, and daemon
  port (57199); the smoke LLM server as the model. Ran an interval
  reminder with netclaw reminder run <id> and confirmed the printed
  result block: Status: ok, the real reply text ("Netclaw smoke
  response."), the session line, and exit 0. Ran it again with
  --json and confirmed the result object, including replyText.
  Pointed the main model at an unreachable provider (the daemon
  picked up the change through its normal config-watch restart),
  ran the same reminder again, and confirmed Status: failed, the
  connection-refused error message, and exit 1. Ran
  netclaw reminder run does-not-exist and confirmed a clear
  not-found message and exit 1. reminder history <id> showed all
  three manual runs with their correct status. All processes stopped
  after the run; the real ~/.netclaw and the default daemon port
  (5199) were never touched.

This PR adds no tests, at the maintainer's request.

Refs #1511. Replaces draft #1527 (stale — see that PR for the original
requirements; not modified here).