Skip building indexes for probers that will only be probed once

CHANGELOG.md

@@ -2,6 +2,7 @@
 
 ## [Unreleased] - ReleaseDate
 
+- Skip building per-subset join indexes for probers that will only be probed once. Join stages now pass a probe-count bound into index acquisition, and single-probe probers answer by scanning the subset directly (a per-trie-node scan budget caps the extra work when a node is probed again later). Most probe-side index builds served exactly one probe, so this removes 60–99% of on-the-fly index-build volume on LLM-compilation benchmarks (whisper −9%, qwen −9%, python_array_optimize −12% end-to-end). A new feature-gated `index-stats` instrumentation (`--features index-stats`) reports per-index build/usage histograms at CLI exit.
 - Speed up query evaluation by building on-the-fly per-subset column indexes as sorted arrays (`SortedColumnIndex`) instead of hash maps. These indexes are typically iterated once and probed a bounded number of times over high-cardinality columns, so skipping hash-table construction is a large win (e.g. ~33% faster on the `gemma` benchmark).
 - Add `make nightly` and `scripts/nightly_bench.py`, a hyperfine-based benchmark harness that measures every `tests/**/*.egg` program at 1/2/4/8 threads and (where supported) in proof-testing mode, caps each run at a 2-minute timeout, skips sub-50ms programs, and emits an HTML dashboard (one row per benchmark, one column per configuration) for nightly.cs.washington.edu. The dashboard uses [eval-live](https://github.com/oflatt/eval-live) for interactive filtering and sorting.
 - Add typed `EGraph` extension state that clones with `EGraph` and is restored by `push`/`pop`.

Cargo.toml

@@ -130,6 +130,8 @@ bin = [
 ]
 serde = ["egraph-serialize/serde"]
 graphviz = ["egraph-serialize/graphviz"]
+# Report join index build/usage statistics at CLI exit.
+index-stats = ["egglog-core-relations/index-stats"]
 
 [dependencies]
 clap = { workspace = true, features = ["derive"], optional = true }

core-relations/Cargo.toml

@@ -49,3 +49,5 @@ harness = false
 [features]
 default = []
 debug-val-trace = []
+# Collect and report statistics on join index builds and their usage.
+index-stats = []