fix(backend): report ingest correctness — propagate storage failure, 413 on oversize, global body cap (closes #23)

A buffer failure on POST /api/v1/reports now returns 500 instead of a
false `ok`, so clients can retry. Decode errors split: an over-limit
body returns 413 (via errors.As on `*http.MaxBytesError`), malformed
JSON stays 400. A new MaxBodyBytes middleware (1 MiB default) caps
every route — rejecting an oversized Content-Length up front and
capping the read otherwise — so the health check and future routes are
bounded too. The raw attacker-controlled geo blob is no longer logged,
only its length; client_id and timestamp are length-bounded before
logging. A decodeJSON handler helper is added. Panic recovery is now a
local middleware routing the stack through slog as structured JSON.
Storage failure uses 500: a full buffer or write error is server-side
and retryable.

Model: opus-4-8

backend/internal/handlers/export_test.go

@@ -0,0 +1,10 @@
+package handlers
+
+import "log/slog"
+
+// NewForTest builds a Handlers around a report sink and logger,
+// bypassing the fx graph so handler behaviour (including the
+// storage failure path) is exercisable in unit tests.
+func NewForTest(buf reportAppender, log *slog.Logger) *Handlers {
+	return &Handlers{buf: buf, log: log}
+}