docs: fix 55 defects found by proofread + source revalidation

[krassx]

A third-party proofread of the docs site produced ~50 findings. Those were revalidated against the working tree, audited a second time adversarially, and then checked against the actual SDK and backend source in the sibling repos under ~/Projects/Bugsee. This PR applies the 55 that survived.

Won't compile as printed

File	Defect
sdk/ios/manual.mdx	Stray comma breaking an ObjC keyword message; extendedReport used where the block parameter is report; Swift if with no braces; ObjC/Swift "equivalent" label samples diverged
sdk/ios/privacy/video.mdx	Swift tab assigned = YES
sdk/ios/custom.mdx	Data(contentsOf:) called without try inside a non-throwing delegate method
sdk/dotnet/installation.mdx	options.monitorNetwork — the property is MonitorNetwork
sdk/android/v6/custom.mdx	Java tab used Report.Type.Crash; on release-6, Report.getType() returns IssueType and no such nested enum exists
sdk/android/v6/maven-installation.md	</manifest missing its closing bracket
integrations/recipes/recipes_issues.mdx	Missing comma made the context sample invalid JavaScript

Wrong facts, settled against source

• migration.mdx sent v6→v7 upgraders to Options.DetectAndReportCrashNdk. It is NdkOptions.DetectAndReport in the bugsee-android-ndk module.
• configuration/ndk.mdx told readers to import com.bugsee.library.ndk.contracts.options.NdkOptions and said the constants "live in the extension's own package". That class is @Deprecated and its Javadoc says the opposite — the keys moved to com.bugsee.library.contracts.options.NdkOptions.
• PerformanceAdaptiveSampling was described three different ways across two pages. Options.java and AdaptiveSampler.java show it promotes transactions that lost the random draw, scored at finish time on status, duration outlier-ness and span-tree complexity. So the static rate is a floor; the documented "ceiling" was backwards on two pages.
• ios/configuration.mdx named the constant BugseeCaptureDeviceAndNetworkNamesKey. BugseeConstants.h declares it without the Key suffix — the one option constant in the header that lacks it.
• captureDeviceAndNetworkNames is genuinely platform-split (iOS NO, Android true). Six wrapper pages each flattened it to a single value repeated across both their iOS and Android tables. Each section now carries its own default. (The proofread reported a .NET-vs-Xamarin contradiction; that pair does not exist.)
• defaultBugPriority was Low in Unity, React Native and Cordova. Android registers IssueSeverity.High and iOS defaults to BugseeSeverityHigh.
• "default log level (normal)" — no such level exists. It is info, per the Flutter enum annotation and Cordova's level || LogLevel.Info. Fixed in 9 files.
• Simulator support was given as v7.0.0 (Cordova) and v9.0.0 (Flutter). It shipped in iOS 6.0.0.
• flutter/release-notes.md displayed 5.5.8 against the 5.5.7 anchor. The anchor was wrong — 8.6.6 already ships 5.5.8, so changing the version text would have invented a regression.
• .NET: duplicate FrameRate row, wrong enum format in the iOS table, stale "1MB" attachment limit, and Xamarin-Studio-era symbolication instructions.
• ios/symbolication.mdx omitted two on-by-default env vars documented elsewhere.
• KMP told readers to pin the version and then showed :+. Pinned to 0.1.1 per version.txt.

⚠️ Retired integrations still documented as working

appserver/config/default.js carries an integrations.retired map. Bitbucket (Cloud removed its Issues REST API on 2026-08-20) and Pivotal Tracker (discontinued 2025-04-30) both throw IntegrationProviderRetiredError on push, but both had full setup guides with no notice and were listed as live on the integrations index. Added retirement notices.

This needs your review — I wrote conservative warnings pointing at Jira / the integrations list, but whether these pages should be redirected or removed outright is a product call.

Deliberately not changed

• http_engine in the Gradle plugin table is correct. Both the proofread and my own first pass called it a casing bug. That column lists registrar keys, and HttpEngineInstrumentation.kt declares override val key = "http_engine". The source comment is explicit: "snake_case is the canonical registrar key; camelCase alias matches the DSL property name."
• android/v6/release-notes.md ordering — a script over all 175 entries found zero violations. The file is version-descending by repo convention.
• compileSDKVersion casing appears in 7 shipped historical release notes. Rewriting published notes is an editorial call, so I left it.
• Android version banners now read 7.x rather than 7.1, so they stop going stale at each point release. Say the word if you'd rather pin them.

Also worth knowing

Nozbe is a live integration (in the providers enum, with both an integrator and an OAuth module) that has no documentation page at all. Its row in the markup-support table is correct and was kept; Hipchat and Testlio were removed, the latter confirmed by migration 056-remove-testlio-integrations.

Findings that did not survive

Four proofread claims and several of my own intermediate conclusions were wrong and are not in this PR — a duplicate Zoho Desk heading, a missing "and" in YouTrack, the .NET/Xamarin default pair, the v6 release-note ordering, and a React Native grammar error that was already correct on main. The MCP tool-count and scope findings were real but had already been fixed on main; only the three AI-plugin pages still claimed the server is read-only.

---

npx cspell clean across 288 files; npm run build succeeds. The two SSG warnings on sdk/dotnet/network and sdk/xamarin/network are pre-existing and untouched.

🤖 Generated with Claude Code