Fix exponential blowup in Pin::Method#combine_same_type_arity_signatures

[apiology]

Fixes #1235.

Problem

Pin::Method#combine_same_type_arity_signatures (lib/solargraph/pin/method.rb) merges same-type-arity overload signatures via reduce + inner flat_map. The flat_map block returns the entire accumulator array from within a block scoped to a single element, so when N existing signatures fail to merge with a new one, the result balloons to N*(N+1) elements instead of N+1. That larger array becomes the accumulator for the next reduce step, so array size compounds multiplicatively with each additional signature — not the O(n^2) the existing guard comment assumes. The guard (return ... if same_type_arity_signatures.length > 10) only checks the initial array length and doesn't catch growth that happens incrementally inside the reduce.

In practice this causes solargraph typecheck --level strong to hang indefinitely with unbounded RSS growth when combining large families of same-type-arity overloads from RBS collection data (e.g. Integer#+, which RBS ships with many dispatch-by-type overloads). See #1235 for full repro steps, rbspy stack evidence (identical stuck stack across repeated samples), and RSS growth data on both macOS ARM64 and Linux x86_64.

Fix

Replace the per-element flat_map (re-emitting the whole accumulator per element) with a single Array#map pass: each element of old_signatures is independently combined-or-kept exactly once, and new_signature is appended only if nothing merged. This preserves the original intent (attempt to merge new_signature into each existing signature; keep unmerged ones as-is) and the intended O(n)-per-signature / O(n^2)-overall cost, without the accidental multiplicative blowup.

Testing

• Added a regression test (spec/pin/method_spec.rb) using doubles that never successfully merge, verifying the combined result length tracks the input length (n) rather than exploding. This test would hang/OOM against the old code.
• bundle exec rspec spec/pin/method_spec.rb: 48 examples, 0 failures (Ruby 3.4, Linux x86_64/Docker).
• Ran the full CI recipe from .github/workflows/typecheck.yml (solargraph_typed job) against this branch on Linux x86_64: SOLARGRAPH_ASSERTS=on bundle exec solargraph typecheck --level strong now completes in ~6.5 minutes (previously hung indefinitely, RSS unbounded). Exit code was 1 with 49 problems found in 14 of 241 files — I believe this is unrelated gem/RBS-version drift from bundle update --pre rbs picking up whatever is latest as of today rather than what was current when v0.59's CI last passed, not a regression introduced by this change (the fix only removes duplicate/exploded array entries; it doesn't change which signatures successfully merge). Flagging this explicitly rather than asserting a clean typecheck, since I haven't diffed against a same-day baseline without this fix (that baseline hangs, by definition, so a like-for-like comparison isn't available).

[apiology]

This PR targeted the v0.59 tag, which has diverged from master significantly (different commit lineages, ~46 commits on master not in v0.59 and 10 commits on v0.59 not in master since their common ancestor) — hence the huge conflict set here.

More importantly, current master already stubbed out combine_same_type_arity_signatures entirely in #1186 as a stopgap for this exact hang ("debug an infinite loop bug in Ruby 3.x"), so the code this PR patches is currently dead on master. Superseding this with a fresh PR based on current master that removes the stub and applies the real fix: see the new PR (linked from #1235).