Leave AssertThrowsOnLastStatement arguments inline when their type can't be rendered

[timtebeek]

What's changed?

AssertThrowsOnLastStatement intermittently crashed with:

Expected a template that would generate exactly one statement to replace one statement, but generated N. Template: #{} #{} = #{any()};

Root cause

• When hoisting a last-statement argument into a preceding local, the recipe renders the argument's type into the declared type of #{} #{} = #{any()};. For an anonymous or local class the type is a fully-attributed JavaType.Class (not JavaType.Unknown), so the Fix AssertThrowsOnLastStatement crash on unattributed argument types #1059 guard (Unknown/None/Void/Null) doesn't catch it. getClassName() then produces invalid Java such as:

MyTest.1 x = __P__.<java.util.concurrent.Callable<java.lang.String>>p();

MyTest.1 doesn't parse as a type, so the stub yields the wrong statement count and the run fails. This only surfaced in real multi-module sources, which is why it was hard to isolate.

Fix

Guard the template apply so any failure leaves the argument inline instead of failing the whole recipe run — a conservative fallback that also covers other un-renderable-type variants. The recipe still moves the earlier lambda statements out; only the un-renderable argument stays inline.

Checklist

• Regression test (argumentWithAnonymousClassTypeIsLeftInline) added and passing

• Full AssertThrowsOnLastStatementTest passes

• licenseFormat clean

• Fixes moderneinc/customer-requests#2861