perf(observable-events): rebuild the generator as a cached 4.8 pipeline

[glennawatson]

What kind of change does this PR introduce?

Refactor and performance, with two fixes: the generator would not load on the .NET 8 SDK, and its generated handlers raised CS8622.

What is the new behavior?

The observable-event generator is an incremental pipeline that actually caches, built against Roslyn 4.8 so every supported SDK can load it.

• It loads on the .NET 8 SDK. Roslyn is pinned to 4.8.0 and the explicit System.Collections.Immutable reference is gone, so the dependency set is whatever the consumer's compiler already carries.
• Generated handlers carry the delegate's own nullable annotations, so a wrapper around EventHandler<T> no longer warns.
  • Gated on the consumer's language version: a C# 7.3 consumer gets no annotations and no #nullable enable, which is source it can compile.
• Nothing but strings crosses the pipeline. No symbol, syntax node, compilation, or Location survives into a cached step; diagnostics and source locations are carried as values and rebuilt when reported.
• Each output is keyed on the smallest model that decides it - one file per host, one per namespace of static events, one for the activation overloads - so a change to one host leaves the rest alone.
• The activation API is an ordinary source output rather than post-initialization output. Post-initialization source is added to the compilation the pipeline then runs against, which makes that compilation new on every run and discards every semantic result cached against the previous one. A single inert post-initialization file costs the entire cache, so the generator emits none.
• Requests are detected without binding to that API, which is no longer visible to the pipeline that produces it.
  • An Events() call that resolves belongs to somebody else and is skipped; one that does not is ours to answer.
  • The static request attribute is matched on how it is written, and the host comes from the typeof argument, which binds on its own.
• Emission runs on a pooled character buffer instead of StringBuilder.

Generated output is unchanged apart from three fixes: <see cref="Foo<T>"/> in a documentation comment was malformed XML and is now <c>Foo&lt;T&gt;</c>, blank lines no longer carry trailing whitespace, and files end with a newline.

What is the current behavior?

• Built against Roslyn 4.14, so the .NET 8 SDK compiler refused to load it (CS9057) and consumers on that SDK got no generated source at all - the build failed on the missing activation API.
• Generated void Handler(object sender, ...) against delegates declaring object?, so every wrapped EventHandler<T> raised CS8622.
• One 1027-line file collected every call site, combined the result with the compilation, and regenerated every file from a single source output whenever anything in the compilation changed.

What might this PR break?

• The package's dependency floor moves to Roslyn 4.8. This widens support rather than narrowing it; nothing that worked before stops working.
• An Events() call that binds to an unrelated method is still skipped, but the test is now inverted. A call that resolves to a method outside the activation class is left alone; a call that resolves to nothing is treated as a request. An ambiguous Events() call (CS0121) resolves to nothing and would now be treated as a request.
• Consumers below C# 8 now get un-annotated source with no #nullable enable. Previously that directive was emitted unconditionally, which is a compile error before C# 8, so this only fixes those consumers.

Checklist

• I have read the Contribute guide
• Tests have been added or updated (for bug fixes / features)
• Docs have been added or updated (for bug fixes / features)
• Changes target the main branch
• PR title follows Conventional Commits

Additional information

Review the pipeline wiring and the two extractors; the rest is mechanical decomposition of the old file.

• EventGenerator.cs - pipeline shape and why the activation API is a source output.
• Helpers/InstanceTargetExtractor.cs and Helpers/StaticTargetExtractor.cs - request detection without the activation API in scope.
• Models/EquatableArray.cs - value equality is the cache key; a defaulted instance and a zero-length one have to compare equal.

A new ReactiveUI.Primitives.ObservableEvents.Benchmarks project drives the generator through CSharpGeneratorDriver over a corpus of 1, 10, and 50 event-bearing hosts, one host per file. Unchanged re-runs a primed driver against the very compilation it was primed on and is the control the incremental cases are read against.

Size	Cold	Unchanged	UnrelatedEdit	EventEdit
1 host	164 us	6.4 us (0.04x)	117 us (0.71x)	148 us (0.90x)
10 hosts	687 us	9.8 us (0.01x)	476 us (0.69x)	527 us (0.77x)
50 hosts	2871 us	26 us (0.009x)	2072 us (0.72x)	2128 us (0.74x)

An edit still costs about 0.72x a cold run and scales with the corpus. Roslyn re-runs every semantic syntax-provider transform whenever the compilation changes at all, and a CPU trace puts roughly 39 percent of that in the compiler binding the Events() call sites. That is intrinsic to keying generation on the receiver type of a call, and no pipeline shape avoids it; removing it would mean keying generation on a declaration instead.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.14%. Comparing base (de69b7f) to head (2971490).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #165      +/-   ##
==========================================
+ Coverage   98.09%   98.14%   +0.04%     
==========================================
  Files         680      703      +23     
  Lines       21359    21733     +374     
  Branches     2622     2679      +57     
==========================================
+ Hits        20953    21329     +376     
  Misses        199      199              
+ Partials      207      205       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.