fix: touchstone empty results table for branches with a slash

[schochastics]

Problem

Touchstone now posts PR comments, but the results table is empty — only the header/footer render (see example).

Root cause

The benchmarks run fine and records are uploaded, but benchmark_analyze() discards all of them with:

Ignoring all benchmarks that don't have exactly those two branches.

The downloaded results artifact shows why — records for a head branch like feat/print-style-modern are stored as:

records/graph_from_data_frame/main
records/graph_from_data_frame/feat/print-style-modern   ← the "/" nests a directory

The receive action installs the {touchstone} R package from its touchstone_ref input, which defaults to @v1. That tag predates the upstream commits "support branch names with slashes" / "encode all filesystem-unsafe characters in branch names" (2025-11-27), so it has no branch_encode(). The raw / in the branch name nests a subdirectory, benchmark_analyze() can't match a benchmark that has exactly the base+head branches, and every benchmark is dropped → empty table.

We previously pinned the action wrapper to a default-branch SHA (7e6072a) to get artifact@v4/v5, but the R package was still installed from @v1. This is the remaining half of that version mismatch.

Fix

Pin touchstone_ref to the same commit the action wrapper already uses, so the installed R package includes branch_encode()/branch_decode().

Only affects R/** etc. via the receive trigger; this PR's own branch (fix-touchstone-slash-branches, no slash) wouldn't have surfaced the bug, so once merged a slash-named branch is the real end-to-end test.

🤖 Generated with Claude Code

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if c4cf214 is merged into main:

• ✔️as_adjacency_matrix: 783ms -> 780ms [-1.48%, +0.74%]
• ✔️as_biadjacency_matrix: 793ms -> 791ms [-1.25%, +0.72%]
• ✔️as_data_frame_both: 1.62ms -> 1.6ms [-3.58%, +0.81%]
• ✔️as_long_data_frame: 4.03ms -> 4.04ms [-2.03%, +2.52%]
• ✔️es_attr_filter: 3ms -> 2.94ms [-5.85%, +2.21%]
• ✔️graph_from_adjacency_matrix: 145ms -> 144ms [-2.07%, +0.16%]
• ❗🐌graph_from_data_frame: 3.6ms -> 3.64ms [+0.23%, +2.18%]
• ✔️vs_attr_filter: 1.73ms -> 1.75ms [-1.47%, +4.12%]
• ✔️vs_by_name: 1.15ms -> 1.16ms [-4.82%, +7.76%]
  Further explanation regarding interpretation and methodology can be found in the documentation.