Perf regression: json_middleware, mTls-handshakes-kestrel, tls-renegotiation-kestrel

| Scenario | Environment | Date | Old RPS | New RPS | Change | Deviation | StDev | Dependencies |
| -------- | ----------- | ---- | ------- | ------- | ------ | --------- | ----- | ------------ |
| json_middleware [Fixed] | Json 5M 14- Trends 28 Core Cobalt Hosted Lin Server | 12/07/2025 11:58:53 | 1,579 | 1,383 | -12.45 % (-197) :thumbsdown: | -3.44 σ | 57.21 | Changes NameVersionDiff |
| mTls-handshakes-kestrel | Kestrel Linux: mTLS Handshakes 14- Trends 28 Core Cobalt Hosted Lin Server | 12/07/2025 12:10:01 | 3,646 | 3,549 | -100.00 % (-97) :thumbsdown: | -2.75 σ | 35.27 | Changes NameVersionDiff |
| tls-renegotiation-kestrel | Kestrel Linux: TLS Renegotiation 14- Trends 28 Core Cobalt Hosted Lin Server | 12/07/2025 12:10:56 | 2,637 | 2,307 | -100.00 % (-330) :thumbsdown: | -12.03 σ | 27.43 | Changes NameVersionDiff |

[PowerBI Dashboard](https://aka.ms/aspnet/benchmarks)

@sebastienros

Start with the benchmark results in this issue and compare the json_middleware, mTls-handshakes-kestrel, and tls-renegotiation-kestrel scenarios in the linked PowerBI Dashboard. Trace the changes and benchmark inputs behind the reported regressions, then rerun the affected scenarios to confirm performance has returned to the expected range.

Written by the indexing model from the issue text.