Never build the P4 object Debug; the measured mode becomes the default

`-Doptimize` defaults to Debug, so the naive `zig build -Dplatform=p4` produced an
object that CANNOT RUN. Debug wraps every tier in `allocators.zig` in a
`DebugAllocator`, whose metadata is page-granular, and one 4 KiB page per size class
does not fit in the 384 KiB the board hands the editor: the image links, flashes, and
then dies in `Pardes.init`. Nothing said so, because every build in this session
happened to pass `-Doptimize=` explicitly.

The p4 target now falls back to ReleaseFast, and that mode was measured rather than
preferred. On the die, against ReleaseSmall over 5 document lengths x 7 trials:

    configuration        fixed    per char   at 160 chars
    ReleaseSmall      16.99 ms     54.3 us       25.56 ms
    ReleaseFast       14.85 ms     34.7 us       20.30 ms   0.79x

13% off the fixed per-keystroke cost, 36% off the per-character cost, for 35% more
flash on a partition that is 39% used. An explicit `-Doptimize=` still wins, so
ReleaseSmall stays one flag away when flash matters more than latency - which is why
this is a fallback and not a hard override.

Following the file's own convention: the web target has pinned ReleaseSmall
unconditionally for the same kind of reason (size is its budget) since before this.

1 files changed, 17 insertions, 2 deletions

diff --git a/build.zig b/build.zig
index 01fed2b6..c9ae800f 100644
--- a/build.zig
+++ b/build.zig
@@ -128,8 +128,23 @@ pub fn build(b: *std.Build) void {
     const enable_jpx = b.option(bool, "jpx", "JPEG 2000 in PDFs, for scanned documents (default on; -Djpx=false drops openjpeg)") orelse true;
     const is_web_target = target.result.cpu.arch == .wasm32 and target.result.os.tag == .freestanding;
     const is_p4_target = target.result.cpu.arch == .riscv32 and target.result.os.tag == .freestanding;
-    // wasm: size is the budget
-    const optimize = if (is_web) .ReleaseSmall else requested_optimize;
+    // wasm: size is the budget.
+    //
+    // p4: Debug is not a supported mode, and `-Doptimize` defaulting to it made the naive
+    // `zig build -Dplatform=p4` produce an object that cannot run. Debug wraps every tier in
+    // `allocators.zig` in a `DebugAllocator`, whose metadata is page-granular; the board hands the
+    // editor a 384 KiB heap and one 4 KiB page per size class does not fit in it, so the image
+    // links and then dies in `Pardes.init`. ReleaseFast rather than ReleaseSmall because it was
+    // measured on the die and not chosen: against ReleaseSmall it is 13% off the fixed
+    // per-keystroke cost and 36% off the per-character cost, for 35% more flash on a partition
+    // that is 39% used. See experiments/report.typ. An explicit `-Doptimize=` still wins, so
+    // ReleaseSmall remains one flag away when flash matters more than latency.
+    const optimize = if (is_web)
+        .ReleaseSmall
+    else if (is_p4_target and requested_optimize == .Debug)
+        .ReleaseFast
+    else
+        requested_optimize;
     // The vendored C is never what we are debugging, and at -O0 it dominates
     // the app: 90% of a Debug startup is tree-sitter's query analyser
     // (perf: ts_query__perform_analysis + ts_lookahead_iterator__next), and