feat(logging): new telemetry design MONGOSH-3408, MONGOSH-3376, MONGOSH-3406

[alenakhineika]

• Replace the Segment analytics integration with a self-hosted fire-and-forget HTTP transport (an intermediate step before integrating with CloudFront). Only TelemetryClient knows about the endpoint URL and HTTP format.
• The previous design sent one analytics event per user action (14 event types); this PR reduces that to three events per session: Identify (device and OS traits), New Connection (connection metadata including topology type, SRV, CSFLE), and Session Ended (aggregated session summary emitted once at close).
• Session data is accumulated incrementally in SessionTelemetryState as the user works, command counts split by REPL vs rc/eval context, error count, snippet count, startup timings, sequence of commands capped at 100, and flushed into a single end-of-session event. The event is only emitted for interactive sessions or identified agentic sessions. Containerized non-interactive non-agentic runs are excluded.
• The internal get-os-info.ts implementation is replaced with the shared @mongodb-js/get-os-info package.
• Use device_id as the sole telemetry user identifier. Remove userId and telemetryAnonymousId (config fields, validators, payload, and the mongosh:new-user/mongosh:update-user bus events).
• Key ThrottledAnalytics cross-session throttle state on device_id instead of anonymousId.
• Add telemetryEndpoint as a CliUserConfig option (with URL validator) carrying the production default. Allow resolving the endpoint as override via MONGOSH_TELEMETRY_ENDPOINT env variable.
• When no endpoint is configured, the telemetry is not sent over the network, but still can be enabled and events are still written to the local log.
• Remove the IS_MONGOSH_EVERGREEN_CI guard and the analyticsOptions/alwaysEnable constructor option (endpoint overrides now via env or config, consistent with updateURL).
• Read user/global config before setting up analytics/telemetry, so the configured telemetry endpoint is honored (safe now that the new-user flow is gone). Defer setup errors until logging is in place.
• Extract analytics wiring into packages/cli-repl/src/setup-analytics.ts (setupTelemetryAnalytics() + AnalyticsOptions), leaving setupAnalytics() a thin wrapper.
• Split the telemetry/logging tests out of cli-repl.spec.ts into new cli-repl-telemetry.spec.ts (fake-server integration), cli-repl-logging.spec.ts (logging config), and setup-analytics.spec.ts (pure-function unit tests); kept log-based unit tests in cli-repl.spec.ts.
• Close each CliRepl instance in a top-level afterEach so per-test resources (driver connections, log writers, process listeners) don't accumulate - the full cli-repl suite now runs within the default heap instead of OOMing.
• Update logging, cli-repl, types, and e2e specs and the cli-repl README for the new telemetry model.

[nbbeeken]

small things, looks really good!