[Console] Prevent concurrent duplicate start/cancel for Flink and Spark apps

[shangeyao]

Problem

Application start/stop checks (isCanBeStart) and state updates (starting()) are not atomic. Rapid duplicate clicks can submit multiple start/cancel requests before startFutureMap/cancelFutureMap is populated.

Proposed solution

• Use in-memory pendingStarts / pendingCancels guards per application id
• Atomically transition DB state to STARTING/CANCELLING via conditional lambdaUpdate
• Clean up pending flags in async completion callbacks and on submission failure

Scope

• FlinkApplicationActionServiceImpl
• SparkApplicationActionServiceImpl

[shangeyao]

Closing because related PR #4385 was closed without merge. Console frontend is being refactored; this work is deferred.