Implement exponential backoff retry logic

[codebestia]

Description

Transient failures (connection resets, 502/503/504 responses) should be automatically retried with exponential backoff and random jitter rather than immediately surfacing to the caller. This makes the SDK resilient to brief network hiccups.

Proposed Steps

• Implement a _retry_with_backoff(fn, max_retries, base_delay) utility.
• Retry on: httpx.ConnectError, httpx.TimeoutException, HTTP 502/503/504.
• Delay formula: min(base_delay * 2**attempt + random.uniform(0, 0.5), 60).
• Do NOT retry on 4xx client errors (except 429, handled separately).
• Log each retry attempt at DEBUG level with attempt number and delay.

Acceptance Criteria

• A request that fails twice then succeeds is returned successfully after the retries.
• A 400 response is NOT retried — it raises InvalidRequestError immediately.
• Retries respect max_retries; exceeding it raises NetworkError.
• The delay between retries grows with each attempt.
• Retry behaviour is testable by injecting a mock transport.

[G-ELM]

I can resolve this task.
Based on my recent PR in the backend, I have a very good idea of the project and i will be able to implement this properly.

[grantfox-oss]

🦊 GrantFox — @G-ELM has been assigned to this issue as part of the Official Campaign campaign!

Next steps:

1. Open a Pull Request referencing this issue (e.g., Closes #10)
2. Your PR will be reviewed by the ShadeProtocol maintainers

Good luck! Track your progress on GrantFox.

[grantfox-oss]

🎉 This issue has been marked as completed on GrantFox as part of the Official Campaign campaign!

@G-ELM's PR #30 was approved and merged by @codebestia.

🏆 @G-ELM: You earned 35 FoxPoints for this contribution! Your current tier: Explorer (118 total points). Track your full progress on GrantFox.

👏 Great work, @G-ELM! Keep contributing to ShadeProtocol.