7.8.2 is not respecting the `_routes` param for data requests with middleware enabled

[rossipedia]

Reproduction

Reproduction repo: https://github.com/rossipedia/rr-data-req-loaders-shenanigans

Instructions are in the README, but I've duplicated them here:

---

Reproduction steps:

Start this application with npm run dev

Execute the following request via curl:

$ curl 'localhost:5173/foo

You should see the following output in the server terminal:

root loader
layout loader
foo loader

Now run the following curl command to send a RR Data request for the routes/foo route only:

$ curl 'localhost:5173/foo.data?_routes=routes%2Ffoo'

You will notice that you still see

root loader
layout loader
foo loader

This is the bug.

When requesting data for a specific set of routes, only those route loaders should execute.

Instead, all the matched route loaders are invoked, based on the request's pathname, and the _routes parameter is ignored.

System Info

System:
    OS: macOS 15.6.1
    CPU: (16) arm64 Apple M4 Max
    Memory: 7.30 GB / 64.00 GB
    Shell: 5.9 - /bin/zsh
  Binaries:
    Node: 22.15.1 - ~/.local/state/fnm_multishells/3342_1756931003739/bin/node
    Yarn: 4.9.1 - ~/.local/state/fnm_multishells/3342_1756931003739/bin/yarn
    npm: 10.9.2 - ~/.local/state/fnm_multishells/3342_1756931003739/bin/npm
    bun: 1.2.21 - /opt/homebrew/bin/bun
  Browsers:
    Chrome: 139.0.7258.155
    Edge: 139.0.3405.119
    Safari: 18.6

Used Package Manager

npm

Expected Behavior

I expect the loaders for only the specified routes to be run.

Actual Behavior

All loaders for all matched routes based on pathname are run, and the _routes param is ignored.

[rossipedia]

This is a bit of a nasty one. It causes a lot of unnecessary loader invocations and downstream reads / API calls that are ultimately discarded / ignored as the data isn't included in the output.

We're also seeing a situation where the response flushes early because it's only waiting for the loaders specified by the _routes parameter, so when the response ends and those other loaders are still in flight, they get aborted. This isn't normally an issue, but it has caused a lot of our downstream infra to deal with a flood of connection-abort errors and similar.

It's also clouding our per-loader metrics as we're seeing a lot more loader calls than we'd expect given the overall HTTP traffic.

I think I know where the issue is and I'm testing out a patch. If it pans out I'll PR the fix.

[brophdawg11]

I think this is the same underlying issue as #14257, so going to close this as a dup for now. Let me know if the fix there doesn't address this.