lib: banner purge + WHY-only comment sweep (rule 8)

Every // ---- section banner dies (132 -> 0): names carry the WHAT.
Narration deleted (filename restatements, run-with lines, what-the-
next-line-does); every ref/hare cite, task cite, divergence, ABI/
layout contract, and ownership qualifier kept (borrowed-view lines
restored where the sweep over-cut). Comment-only proven: all 442
walk-workdir .s and 32 import-probe .s byte-identical before/after;
libbyteid 56-roster all-ID.

lib/strings/tokenize_test.ww

@@ -162,7 +162,6 @@ fn expect_str_done(t: *strings.tokenizer) void = {
 	assert(!(!streq(strings.remainingtokens(&t2), "a b c")));
 };
 
-// ---- splitn / rsplitn / split ----------------------------------------
 // ref/hare/strings/tokenize.ha:245 @test fn split. Hare's vectors
 // mirrored here directly; element reads go through `&toks.ptr[i]: *str`
 // rather than `toks[i]` so the 16B str element copy stays out of the
@@ -290,7 +289,6 @@ fn expect_str(toks: []str, i: i32, want: str) void = {
 	os.free(t3.ptr: *void, (t3.cap: u64) * size(str): u64);
 };
 
-// ---- cut / rcut -------------------------------------------------------
 // ref/hare/strings/tokenize.ha:316 (@test fn cut). str wrappers over
 // bytes.cut/rcut. The `let (a, b) = cut(...)` destructure drives the
 // over-cap tuple-return (sret) path end-to-end (a 2nd #10 witness).