Improve overload resolution and macro handling in Chain::Call

[apiology]

Fixes macro-based return type inference when the macro's $1/$2 placeholders reference the call's actual arguments, not just literal text elsewhere:

class Foo
  # @!macro
  #   @return [$1]
  def self.bar; end
end
Foo.bar(String)
#        ^^^^^^ should substitute into the return type

	Before	After
Foo.bar(String) inferred type	nil	String

Root cause: macro/directive reprocessing only ran when normal overload matching produced no type at all. bar's macro-based signature did produce a (wrong) type through the plain matching path, so reprocessing never ran. Chain::Call#inferred_pins's per-overload matching logic is extracted into match_overload_type, restructured so macro reprocessing runs whenever the matched type doesn't actually resolve the macro's placeholders against the real arguments.

Also adds spec coverage for related overload/macro edge cases; 8 of the new specs are pending, linked to tracking issues rather than left silently broken (#1223, #1246).

Testing: bundle exec rspec spec/source/chain/call_spec.rb spec/source_map/clip_spec.rb — 203 examples, 0 failures, 31 pending.

[apiology]

@lekemula @castwide: Ready for review

[apiology]

If there's a PR that a spec is pending on, just put the PR URL, no other narrative.

[apiology]

Claude:

Done in 59996fa - all 6 occurrences of this pending reason simplified
to the bare URL.

[apiology]

Are the two headlines from this PR (overload resolution and macro handling) dependent on each other? What would be the trade-offs of moving the overload code over to 1223?

[apiology]

Claude: No, not dependent within this PR. match_overload_type (the overload-resolution extraction, pure lift of the existing inline loop body into a method, no behavior change) and process_macro/process_directive (the new macro/directive reprocessing, added after that loop, gated on p.macros/p.directives) touch different points in inferred_pins and never call into each other. Reverting one wouldn't remove the other.

The real coupling is with #1223, not between this PR's own two headlines. #1223 independently rewrites the same sorted_overloads.each loop in the same method (wrapping it in a two-pass require_literal retry for literal-overload matching), and touches the identical three files this PR touches: lib/solargraph/source/chain/call.rb, spec/source/chain/call_spec.rb, spec/source_map/clip_spec.rb. Whichever PR merges second has to hand-reconcile the loop body against the other's restructuring.

Trade-offs of moving the overload-resolution half of this PR into #1223:

• Pro: the reconciliation happens once, by whoever's already deep in that loop for Infer element types for array literals #1223, instead of being redone independently after Infer element types for array literals #1223 merges.
• Con: Infer element types for array literals #1223 already touches 26 files; folding this in grows it further, and ties the macro fix's landing to Infer element types for array literals #1223's review timeline even though process_macro/process_directive have no overlap with Infer element types for array literals #1223's changes at all.

I'd leave this PR as-is and let whichever of the two merges second do the reconciliation - the overlap is confined to one loop, not the whole overload-matching mechanism.