feat: label SessionTime histogram by process type

[GregTheGreek]

Summary

`relayer.SessionTime` has always lumped all three TSS workloads into one distribution, which makes its quantiles meaningless:

Workload	Typical duration	Dominant cost
Keygen	tens of seconds	Paillier key generation
Signing	hundreds of ms	CMP rounds + libp2p
Resharing	seconds	reshared key derivation

Any dashboard p50/p95 computed over this histogram is a weighted average of three wildly different curves.

Change

• Added `Type() string` to `TssProcess` (returns `"keygen"`, `"signing"`, `"resharing"`)
• Widened `Metrics.StartProcess` to `StartProcess(sessionID, processType string)`
• `MpcMetrics` now stores `{at, processType}` per session and emits the `type` attribute on `SessionTime.Record`
• Regenerated coordinator mock; updated `CoordinatorTestSuite.SetupTest` expectation

No change to cache.Metrics or the `EndProcess` signature - the type is recovered from the stored session entry.

Grafana note

After this lands, query p95 as:

```
histogram_quantile(0.95, sum by (le) (rate(relayer_SessionTime_bucket{type="signing"}[5m])))
```

Dashboards that don't filter on `type` will silently aggregate all three workloads, same as today.

Test plan

• `go build ./...` clean
• Go tests in affected packages pass locally
• CI green
• Staging: confirm `SessionTime` metric shows `type` label in the collector

Notes for reviewers

• `EndProcess` only emits the type present at `StartProcess`. If you ever call `EndProcess` for a session we didn't start, the emit is skipped (existing warning path).
• Process types are plain strings rather than a typed enum to keep cardinality obvious and avoid a cross-package dep for consumers.

[github-actions]

Go Test coverage is 53.3 %\ ✨ ✨ ✨

[mpetrun5]

The other workloads basically never happen so I don't think for the purposes of metrics it does anything.