Add declarative resilience primitives to DSL: retry/backoff/on_error, timeout, sub-second sleep, and set-style fan-out

[Copilot]

The DSL lacked first-class resilience and fan-out constructs, forcing users to hand-build retry/backoff/error flows with loop/if/sleep and manual coordination. This change adds declarative primitives for retry, timeout, error handling, sub-second timers, and array-based parallel composition.

• DSL resilience primitives

  • Added df.retry(...) with policy, max attempts, exponential backoff controls, jitter, and optional on_error handler.
  • Added df.catch(fut, handler) and df.on_error(fut, handler) as explicit failure-branch combinators.
  • Added df.timeout(fut, seconds) for first-class deadline wrapping.

• Sub-second durable sleep

  • Added overloaded df.sleep(double precision) and updated df.sleep(bigint) encoding to milliseconds.
  • Runtime sleep execution now supports both legacy integer-second payloads and new millisecond config for backward compatibility.

• Set-style fan-out

  • Added df.parallel(text[]) and df.join_all(text[]) to compose N-way joins without pairwise join/join3 chaining.

• Runtime orchestration behavior

  • Implemented retry execution semantics in loop handling:
    • policy-based retryability checks,
    • capped exponential backoff with deterministic jitter,
    • terminal fallback branch execution via on_error_node.
  • Implemented timeout wrapper handling in race execution to return deterministic timeout failures.

• Graph/config handling + grants/docs

  • Extended config child traversal/transformation to include on_error_node so nested handler nodes are validated and materialized correctly.
  • Updated function grant allowlists to include the new DSL signatures.
  • Updated user guide DSL reference for the new primitives.

SELECT df.start(
  df.retry(
    df.timeout(
      df.sql('SELECT call_external_api()'),
      0.5
    ),
    policy            => 'on_429',
    max_attempts      => 10,
    initial_backoff_ms=> 200,
    max_backoff_ms    => 30000,
    backoff_multiplier=> 2.0,
    jitter            => 0.2,
    on_error          => df.sql('SELECT log_terminal_failure()')
  )
);