Extend Application Metrics retention beyond 30 days

## Problem

Application Metrics are currently retained for **30 days across all plans** (Developer, Team, Business, Enterprise). This makes month-over-month (MoM) trend analysis impossible natively in Sentry — users cannot compare this month's metrics to last month's without exporting data to an external system.

## Gap

- 30-day retention is sufficient for short-term debugging but insufficient for performance trending, regression detection across releases, or any comparative analysis that spans billing cycles.
- Users needing MoM or QoQ analysis must maintain a separate metrics store alongside Sentry, undermining the value of Application Metrics as a consolidated observability layer.
- Sentry already offers **13 months of downsampled retention for span data** on Business/Enterprise plans. No equivalent exists for Application Metrics.

## Customer signal

Multiple customers have raised this. Notably, at least one customer explicitly stated they are comfortable with **cold or downsampled storage** for data beyond 30 days — retrieval speed is not a requirement. This suggests a sampled/tiered retention model (similar to spans) would satisfy the use case at lower infrastructure cost.

## Options

- **Tiered retention**: offer full-fidelity data for 30 days + downsampled/aggregated retention for 90 days or 13 months on higher plans, mirroring the spans sampled retention model.
- **Extended flat retention**: increase the retention window to 90 days on Business/Enterprise plans.
- **Export / data pipeline**: allow periodic export of metric rollups so users can maintain their own historical store — lower lift, partial solution.

--

[View Junior Session in Sentry](https://sentry.sentry.io/explore/conversations/slack%3AD0ANAH23ZV4%3A1783349303.510569/?project=4510944073809921)

No files, tests, or entry points are named. Start by locating Application Metrics retention configuration and the existing span downsampling or retention model, then compare the available options and their plan limits. Done would require an agreed retention policy implemented with coverage for the affected plans and retention behavior.

Written by the indexing model from the issue text.