feature: diagnostic report collector for issue reporting (#65)

[mcpolo99]

Summary

Implements the diagnostic report collector from #65. When obfuscation fails (or succeeds), users can now produce a complete, copy-pasteable markdown report with one CLI flag or one GUI button — system info, project configuration, run result, and the full log transcript.

Fixes #65

What was added

Core — Confuser.Core/Diagnostics/

• DiagnosticCollector — decorates both ILogger and IProgressReporter. Passes every call through to the real logger/reporter while capturing a full-verbosity transcript, timing, and outcome.
• DiagnosticReport — renders the captured data into a self-contained markdown report (System / Project Configuration / Result / Log Output / Elapsed). Never throws.
• DiagnosticRedactor — scrubs the user-profile path (%USER%) from text destined for a public report.

CLI — --dump / --dump=<file> flag. Wraps the logger+reporter, writes the report after the run (success or failure), prints the path.

GUI — "Copy Report" button on the Protect tab, enabled once a run completes; copies the report to the clipboard.

Design notes — caveats and how they're handled

The issue's own "Caveats" section only named memory usage; hooking the logging core and pasting output publicly raised several sharper edges, all addressed:

• Secret / PII leakage (report is pasted publicly): strong-name passwords (SNKeyPassword, SNSigKeyPassword) are never emitted; absolute paths have the user-profile prefix replaced with %USER%, applied to both the config block and captured log lines (the engine's PrintEnvironmentInfo dumps assembly locations into the log stream too).
• Finish() moved to IProgressReporter (in the Replace custom ILogger with Serilog + Microsoft.Extensions.Logging #64 logging migration, so the issue text is stale): the collector decorates the progress reporter as well, giving it the authoritative success flag + timing.
• Capture vs. display verbosity: capture is independent of the inner logger's level (IsEnabled → true, entries always retained), so a report from a default Information run still contains the Debug detail needed to diagnose. Display filtering is preserved by only forwarding to the inner logger when it's enabled for that level.
• Packer nested runs: Finish is last-wins, so the top-level run — which finishes last — determines the reported outcome/elapsed.
• Thread safety: bounded ring buffer (last 2000 entries) behind a lock, with a dropped-entry count surfaced in the report so truncation is never silent.
• Markdown break-out: dynamic code-fence longer than any backtick run in the log body.
• Robustness: report generation is fully defensive — it runs precisely when things are already broken.

Known limitation

The sample report in the issue shows a "Target Framework" line, but ConfuserProject has no TFM field, so it's omitted for now (the log output already shows the auto-detected runtime paths and the module). Could be added later as a best-effort read of the module's TargetFrameworkAttribute.

Testing

• 19 Core unit tests (collector capture/forwarding, full-verbosity, ring-buffer + dropped count, last-wins Finish, redaction, dynamic fence, never-throws).
• CLI E2E test — runs the CLI with --dump against the pre-built sample and asserts the report is written with the expected sections and no secret leakage.
• Full offline CI (local-ci.sh): MSBuild build across all TFMs + entire test suite — all green.

Example report

## System
- OS: Microsoft Windows 10.0.26100
- Runtime: .NET 10.0.9
- ConfuserExx: Confuser.Core 1.7.0-alpha.28

## Project Configuration
- Base Directory: %USER%\...\dumptest
- Modules: SampleApp.exe
- Protections: rename
- Packer: (none)

## Result: SUCCESS

## Log Output
... full Debug transcript ...

## Elapsed: 0.9 s

[github-actions]

Lint Results

✅ All checks passed

Check	Result
Whitespace (indentation, line endings)	✅
Style (IDE rules, naming, var usage)	✅
Analyzers (CA*, RCS* rules)	✅