FileIO: Configurable HTTP retry policy for Reader/Writer

[drslebedev]

Add an optional retry policy for HTTP(S) operations in fileio so transient network errors don’t immediately fail a Reader / Writer.

Motivation

Right now file read/write and HTTP GET/POST in FileIO are “one shot”: any temporary error (timeouts, 5xx, connection reset, etc.) aborts the operation. For long-running or unstable links, a built-in retry mechanism would make the API much more robust.

Proposed changes

• Extend ReaderConfig / WriterConfig with a RetryPolicy:

  enum class BackoffStrategy { none, linear, exponential };
  
  struct RetryPolicy {
      std::size_t maxAttempts      = 1;      // 1 = no retry (current behavior)
      std::uint64_t initialDelayNs = 0;      // delay before first retry
      std::uint64_t maxDelayNs     = 0;      // backoff
      BackoffStrategy backoff      = BackoffStrategy::none;
      std::vector<long> retryableStatusCodes; // e.g. {408, 429, 500, 502, 503, 504}
      std::function<bool(long httpStatus, cpr::ErrorCode ec)> shouldRetry; // optional override
  };

• Add std::optional<RetryPolicy> retry; to ReaderConfig and WriterConfig.

• Implement retry loop in

• Default behavior: maxAttempts = 1 ⇒ no retries (preserve current semantics).

Notes / Open questions

• How to combine with longPolling for GET? (Probably no retries there, or retries only on initial connect.)
• Emscripten: respect main-thread constraints while sleeping between retries.