p2: digest with in-sift FIFO, SQLite caches, Honcho context, recover

Dixie is capped (max_inflight 2, queue_depth 8, queue_wait 20s) and
never queued on llama-server. Streaming chat, thinking off, 45s
fallback to clean. Digest and chunk-embedding caches are
content-addressed SQLite (Honcho context is not cached). Recover
uses the guarded Fetcher. guardian-star is refused in code.

vendor/github.com/dustin/go-humanize/big.go

@@ -0,0 +1,31 @@
+package humanize
+
+import (
+	"math/big"
+)
+
+// order of magnitude (to a max order)
+func oomm(n, b *big.Int, maxmag int) (float64, int) {
+	mag := 0
+	m := &big.Int{}
+	for n.Cmp(b) >= 0 {
+		n.DivMod(n, b, m)
+		mag++
+		if mag == maxmag && maxmag >= 0 {
+			break
+		}
+	}
+	return float64(n.Int64()) + (float64(m.Int64()) / float64(b.Int64())), mag
+}
+
+// total order of magnitude
+// (same as above, but with no upper limit)
+func oom(n, b *big.Int) (float64, int) {
+	mag := 0
+	m := &big.Int{}
+	for n.Cmp(b) >= 0 {
+		n.DivMod(n, b, m)
+		mag++
+	}
+	return float64(n.Int64()) + (float64(m.Int64()) / float64(b.Int64())), mag
+}