fix(tlsmiddlebox): make ClientId settable and validate its value - #1811

Checklist

• I have read the contribution guidelines
• reference issue for this pull request: none, found this while poking at experiment options
• if you changed anything related to how experiments work and you need to reflect these changes in the ooni/spec repository, please link to the related ooni/spec pull request: N/A, no data format change
• if you changed code inside an experiment, make sure you bump its version number: 0.1.2 to 0.1.3

Description

tlsmiddlebox declares its ClientId option as int, but SetOptionAny in internal/registry only knows how to set int64, bool, and string fields. So the option is documented (it has an ooni tag and shows up in the options list) but you cannot actually use it. Applying the registry test this PR adds on top of current master:

$ go test -count=1 -run 'TestExperimentOptionsAreAlwaysSettable' ./internal/registry/
--- FAIL: TestExperimentOptionsAreAlwaysSettable (0.00s)
    --- FAIL: TestExperimentOptionsAreAlwaysSettable/tlsmiddlebox (0.00s)
        factory_test.go:1138: field ClientId has kind int, which SetOptionAny cannot set

miniooni tlsmiddlebox -O ClientId=2 is where that bites. miniooni passes option values as strings, string values for integer options go through setOptionInt, and SetOptionAny only routes to setOptionInt for fields whose kind is int64. An int field falls through to the default branch instead, which returns unsupported option type: string. Every other experiment already declares integer options as int64; tlsmiddlebox was the only one using plain int.

There is a second problem behind the first one. OONI Run v2 descriptors apply options with SetOptionsJSON, which unmarshals straight into the config struct and bypasses the kind check, so a descriptor can set any ClientId it likes. ClientIDs in tracing.go is a map[int]*utls.ClientHelloID with keys 1 to 4, the lookup silently returns nil for any other key, and netxlite.NewUTLSConn dereferences the pointer. Running the measurer with Config{ClientId: 5} against a local TLS listener on current master:

panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x0 pc=0xb5ca41]

github.com/ooni/probe-cli/v3/internal/netxlite.NewUTLSConn(...)
	internal/netxlite/utls.go:83
...
github.com/ooni/probe-cli/v3/internal/experiment/tlsmiddlebox.(*Measurer).handshakeWithTTL(...)
	internal/experiment/tlsmiddlebox/tracing.go:100

The fix is small: ClientId becomes int64 (with clientid() casting for the map lookup, so tracing.go is untouched), and Run now rejects a ClientId that does not match any known fingerprint, the same way it already rejects an invalid test helper. I picked an error over a silent fallback to the stdlib handshaker because somebody who asked for a Chrome fingerprint should not quietly measure with the wrong ClientHello.

The registry test shown above walks every experiment config and flags any ooni-tagged field whose kind SetOptionAny cannot set, so a future experiment cannot reintroduce this class of bug.

Testing

$ go test -count=1 -race ./internal/experiment/tlsmiddlebox/ ./internal/registry/
ok  	github.com/ooni/probe-cli/v3/internal/experiment/tlsmiddlebox	61.851s
ok  	github.com/ooni/probe-cli/v3/internal/registry	1.084s

New coverage: the invalid ClientId path in Run, the non-default clientid() value, and the registry-wide option kind check. go vet and gofmt are clean on both packages.