Sites return HTTP 500 with a correct response body above ~36 concurrent connec

### 👟 Reproduction steps Every site on a self-hosted 2.0.0 instance starts returning `500` once a client opens more than roughly 36 **simultaneous TCP connections** to it. The failures are not random: the count is identical across repeated runs. The striking part is the shape of the failing response. **The body is byte-for-byte the correct file** — only the status line and `Content-Type` are wrong. ```bash # 60 requests to the same site, 40 connections in parallel python3 -c " import random for i in range(60): print('-o'); print('/dev/null') print('https://YOUR-SITE/?p=%d%d' % (random.randint(1000,9999), i)) " > args.txt tr '\n' '\0' < args.txt | xargs -0 curl -s -Z --parallel-max 40 -w '%{http_code}\n' \ | sort | uniq -c ``` Result on our instance, three consecutive runs, two unrelated sites: ``` 55 200 5 500 ``` To see the malformed response itself: ```bash curl -sD headers.txt -o body.bin "https://YOUR-SITE/_app/immutable/chunks/SOME-CHUNK.js" # repeat under parallel load ``` ``` HTTP/2 500 content-type: text/html; charset=UTF-8 cache-control: no-cache, no-store, must-revalidate server: swoole-http-server import{Q as h,A as w,R as k,I as x,...}from"./OTHER-CHUNK.js";function n(r,a){... ``` The body is valid JavaScript, `21470` bytes, and `cmp` reports it identical to the file the same URL returns when the request succeeds. Only the metadata is wrong. ### 👍 Expected behavior Either the request succeeds with `200` and `application/javascript`, or — if the server is genuinely out of capacity — it queues the request, or rejects it with a status that says so (`503`, `429`) and a matching body. A `500` carrying the correct payload under `Content-Type: text/html` is the worst of both: browsers refuse to execute the module, and nothing in the response indicates overload. This also appears to contradict the stated design of Hyperloop B, which is described as decoupling concurrency from process count and removing "the artificial ceiling imposed by worker poo…