fix: Make the FDv1 fallback TTL-based and never indefinite

[beekld]

Requirements

• I have added test coverage for new or changed functionality
• I have followed the repository's pull request submission guidelines
• I have validated my changes against all supported platform versions

Related issues

SDK-2984 (part of SDK-2935). Implements the SDK-2527 spec change (sdk-specs#251, DATASYSTEM v2 1.6 / CSFDV2 8.3). Reference: FDv1FallbackDirective::DefaultTtl / FromServiceTtl in cpp-sdks, cpp-sdks#631.

Describe the solution you've provided

Before this change, a TTL of 0 meant "fall back indefinitely". The header and goodbye.protocolFallbackTTL TTLs were used without validation, the 1 hour default had no jitter, and FDv2 was never retried after the FDv1 fallback synchronizer was engaged.

• resolveFallbackTtl (fallback_directive.dart): a TTL in (0, 1h] is used as-is. A missing, malformed, zero, negative, or > 1h TTL becomes 1h - uniform[0, 1800]s. This covers both the x-ld-fd-fallback-ttl header and goodbye.protocolFallbackTTL, which already reach the orchestrator as FDv2SourceResult.fdv1FallbackTtl.
• Every directive goes through FDv2DataSourceOrchestrator._processDirective, whether or not an FDv1 tier is configured. It resolves the TTL, calls SourceManager.engageFdv1Fallback(), and schedules recovery. engageFdv1Fallback() blocks the FDv2 synchronizers and enables the FDv1 slot if there is one. With no FDv1 slot, nothing is available, so the synchronizer loop waits for recovery.
• When the TTL elapses, SourceManager.restoreFdv2Synchronizers() makes the FDv2 slots available again and blocks FDv1. The active FDv1 run then ends (recover), so FDv1 is closed before the FDv2 synchronizer is created and the two never run together. If the FDv1 tier fails first, the loop waits for the timer instead of halting.
• Initializers still run only once, in order. A directive from an initializer ends the initializer phase, and the remaining initializers don't run. When no FDv1 tier exists to complete initialization, any data the initializers already delivered completes it.
• Removed _DirectiveAction.deferRetry and _pendingRetryDelay; the no-tier case now uses the same path as the FDv1 tier.
• Updated comments that described Duration.zero as indefinite.

Describe alternatives you've considered

I kept the raw TTL as Duration? through the protocol and source layers and resolved it once in the orchestrator. Validating it in each parser would have meant changing every parser; this way the parsers are unchanged and the rule lives in one place.

Additional context

• Deliberately unchanged: contract tests. The existing harness has no fallback-TTL coverage, so only unit tests were added.
• fallbackTtlResolver is injectable on the orchestrator so tests can scale the TTLs down and keep real waits short.

Link to Devin session: https://app.devin.ai/sessions/f4e0848a068c4648b778346105ba7792
Open in Devin Desktop: https://app.devin.ai/desktop/session/f4e0848a068c4648b778346105ba7792?variant=devin
Requested by: @beekld

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring