Auto merge of #140168 - joshlf:no-nfas, r=jswrenn

transmute: Mark edges by byte sets, not byte values

This leads to drastic performance improvements. For example, on the author's 2024 MacBook Pro, the time to convert the `Tree` representation of a `u64` to its equivalent DFA representation drops from ~8.5ms to ~1us, a reduction of ~8,500x. See `bench_dfa_from_tree`.

Similarly, the time to execute a transmutability query from `u64` to `u64` drops from ~35us to ~1.7us, a reduction of ~20x. See `bench_transmute`.

r? `@jswrenn`

Author: bors

Cargo.lock

@@ -4521,6 +4521,7 @@ dependencies = [
  "rustc_hir",
  "rustc_middle",
  "rustc_span",
+ "smallvec",
  "tracing",
 ]
 

compiler/rustc_transmute/Cargo.toml

@@ -5,11 +5,13 @@ edition = "2024"
 
 [dependencies]
 # tidy-alphabetical-start
+itertools = "0.12"
 rustc_abi = { path = "../rustc_abi", optional = true }
 rustc_data_structures = { path = "../rustc_data_structures" }
 rustc_hir = { path = "../rustc_hir", optional = true }
 rustc_middle = { path = "../rustc_middle", optional = true }
 rustc_span = { path = "../rustc_span", optional = true }
+smallvec = "1.8.1"
 tracing = "0.1"
 # tidy-alphabetical-end
 
@@ -20,8 +22,3 @@ rustc = [
     "dep:rustc_middle",
     "dep:rustc_span",
 ]
-
-[dev-dependencies]
-# tidy-alphabetical-start
-itertools = "0.12"
-# tidy-alphabetical-end