perf: two-lifetime arenas + streaming XML/JSON output

.gitignore

@@ -17,10 +17,16 @@ evtx-wasm/evtx-viewer/public/pkg
 profile/*
 binaries/*
 benchmarks/*
+!benchmarks/omer-pc_ablation_matrix_t1_20251227.json
+scripts/FlameGraph
 .PRE_PATH
 
 # Local-only fixtures (e.g. PE/CRIM blobs for WEVT_TEMPLATE work). Keep out of git.
 samples_local/
 
 # Local vendor checkouts (for format research / patching upstream).
 external/
+profile_results/*
+
+# Local scratch space (worktrees, ablation builds, etc).
+tmp/

Cargo.toml

@@ -40,6 +40,9 @@ tempfile = { version = "3.3.0", optional = true }
 
 serde = "1"
 serde_json = { version = "1", features = ["preserve_order"]}
+bumpalo = { version = "3", features = ["collections"] }
+itoa = "1"
+ryu = "1"
 
 [target.'cfg(not(windows))'.dependencies]
 # jemalloc is significantly more peformant than the system allocator.
@@ -58,6 +61,8 @@ multithreading = ["rayon"]
 # Enable WEVT_TEMPLATE extraction helpers (used by `evtx_dump extract-wevt-templates`).
 wevt_templates = ["glob", "goblin"]
 
+# Perf ablation toggles (used for local benchmarking; no effect unless enabled).
+
 [dev-dependencies]
 insta = { version = "1", features = ["json"] }
 pretty_assertions = "1.2.1"