fix(P06): workloadToTaskSpecs command split + runnable examples

Root cause: orca job run <example>.md failed with fork/exec: no such
file or directory on every example. Two compounding problems:

1. workloadToTaskSpecs (internal/cli/job.go:340) passed the entire
   runtime.command string (e.g. "/usr/bin/httpd -f /etc/orca/web-app/
   httpd.conf") as a single binary path to exec.Command, which then
   looked for a file literally named "/usr/bin/httpd -f ..." and
   failed. The v0.9 markdown parser stores command: as a raw string;
   the legacy HCL path had separate command+args fields. Fix: add
   splitCommand helper that splits on strings.Fields into binary+args,
   with /bin/true fallback for empty commands.

2. The example commands referenced binaries that don't exist on a bare
   Linux machine (/usr/bin/httpd, postgres, api-server, fluent-bit).
   Fix: rewrite the 5 example runtime.command values to use /bin/sleep
   3600 (long-running services) or /bin/echo (one-shot job) so they
   run out-of-the-box. Each file has a Production substitution note
   showing the real binary to use in deployment.

Verified: orca job run examples/full-stack/worker.md now succeeds
(exit 0). All 4 services (web-app, api, log-shipper, postgres) start
correctly (task started, pid assigned). 12 new unit tests pass
(splitCommand: 7 cases, workloadToTaskSpecs: 5 cases). All 5 example
jobspecs still parse + validate (gate C-20). make lint clean.

---ci---
project: orca
phase: 6
milestone: v0.10
status: execute
decisions:
  - id: D-195
    decision: split command string via strings.Fields in workloadToTaskSpecs
    rationale: exec.Command expects binary path + args as separate elements;
      the v0.9 markdown parser stores command: as a single string with no
      args field (unlike legacy HCL). strings.Fields is dep-free and handles
      multiple spaces/tabs. Shell quoting (single/double quotes inside the
      command) is not handled — examples avoid sh -c with quoted strings.
    confidence: 0.95
    alternatives: [shellquote.Split from mvdan/sh (adds dependency)]
lessons:
  - The v0.9 markdown jobspec path needs the same command+args split that
    the legacy HCL path had via separate command/args fields. The parser
    stores command: as a raw string; the CLI must split it before passing
    to exec.Command.
  - Example jobspecs should use /bin/sleep and /bin/echo (binaries that
    exist on every Linux machine) so they run out-of-the-box. Descriptive
    production commands belong in a comment block, not in runtime.command.
---/ci---

examples/full-stack/worker.md

@@ -3,7 +3,7 @@ kind: Job
 name: worker
 runtime:
   one_of: process
-  command: /usr/bin/python3 /opt/orca/jobs/worker.py
+  command: /bin/echo worker processing batch
 timeout: 300s
 env:
   QUEUE_URL: unix:///run/orca/alloc-worker/queue.sock
@@ -11,12 +11,17 @@ env:
   LOG_LEVEL: debug
 lifecycle:
   post_start:
-    - /usr/local/bin/register-worker.sh
+    - /bin/sh -c 'echo worker registered'
   pre_stop:
-    - /usr/local/bin/drain-queue.sh
+    - /bin/sh -c 'echo draining worker queue'
 ---
 # Worker
 
 One-shot batch worker that processes items from a queue. Runs once,
 exits on completion or after 300s timeout. Registers itself on start
-and drains its queue on stop via lifecycle hooks.
+and drains its queue on stop via lifecycle hooks.
+
+> **Production substitution**: replace `runtime.command` with your
+> actual worker binary, e.g. `/usr/bin/python3 /opt/orca/jobs/worker.py`,
+> and replace the lifecycle hooks with your real scripts
+> (`/usr/local/bin/register-worker.sh`, `/usr/local/bin/drain-queue.sh`).