Add condaEnvCount and managerCount to pet.refresh telemetry

[eleanorjboyd]

Summary

Adds two environment-shape measures to the pet.refresh telemetry event:

• condaEnvCount — number of discovered environments whose kind is Conda
• managerCount — number of discovered environment managers

The event already carries envCount (total discovered environments) and Duration. Together with the two new measures, we can slice refresh duration by total env count and conda footprint in telemetry, directly testing the open hypothesis that the slow-refresh cohort is dominated by conda-heavy / many-env setups.

Motivation

We have an unexplained elevation in bad-experience rates that our simpler hypotheses (experiment ramp, first-run cost, PET code regression) did not account for. The remaining signal points at engaged users with complex setups, suspected conda-heavy — but we can't confirm this per-session without instrumenting the shape of what PET discovers. This is a minimal instrumentation change to unblock a confident decision (and to later verify whether lazy-Conda helps the right sub-cohort).

Notes

• No new events; both new fields are measurements on an existing event.