fix(#30): detect Cloudflare challenge pages and retry with premium_proxy

[Jing-yilin]

Problem

Closes #30

When Cloudflare serves a JS challenge interstitial page, ScrapingBee returns HTTP 200 with a small challenge HTML (no real project content). The existing retry logic only triggers on HTTP 429/5xx, so challenge pages passed through silently with 0 parsed campaigns. The nightly crawl sanity check then failed with:

ERROR: crawl sanity check FAILED — only 0 distinct campaigns seen (expected >=50)

Root Cause

CF challenge pages are small (<200KB) and contain challenge-platform/scripts/jsd/main.js. Real Kickstarter discover pages are 500KB+ and contain data-project elements. The two are easy to distinguish by size + marker.

Fix

scrapingbee_client.go

• Added isCFChallengePage(html string) bool — detects challenge pages by the presence of CF's challenge-platform script in a small (<200KB) response
• In doRequest, if a 200 response is a CF challenge, it's treated as a retryable error (logged + continue). The existing retry loop will escalate to premium_proxy=true on the 4th attempt
• Added response size to success log for easier debugging

kickstarter_parser.go

• Added warning log when 0 campaigns are parsed from a large (>50KB) page — surfaces HTML structure changes early

scrapingbee_client_test.go

• 5 unit tests covering pure challenge, large real page with CF script injected, normal page, empty, small non-CF page

[Jing-yilin]

Finding:

• Medium: backend/internal/service/kickstarter_parser.go:49 This new warning is in the shared parser, so it now fires for every caller that legitimately gets zero campaigns back, not just the nightly crawl. Search() can return empty results for a user query, and the crawl itself uses len(campaigns) == 0 as the normal pagination stop condition. On any of those pages that happen to be larger than 50 KB, we will now log possible HTML structure change or blocked response for an expected outcome, which dilutes the signal this change is trying to add. I would move this warning to the crawl call site or gate it on a stronger blocked-response signal instead of any large 0-result page.