[Performance] Leader Election Slow - No Caching of Cluster State

## Description
Leader election is slow due to missing caching of cluster state. Every election queries database for full cluster state without caching, causing unnecessary latency during leadership transitions.

## Steps to Reproduce
1. PD cluster detects leader failure
2. Initiates leader election process
3. Queries full cluster state for each candidate
4. No caching of state information
5. Election takes 2-3 seconds instead of <100ms

## Environment Information
- System: PD leader election
- Query: Cluster topology and state
- Caching: Not implemented
- Frequency: On every election

## Expected Behavior
- Cluster state cached locally
- Cache invalidated on topology changes
- Leader election <100ms
- Background cache refresh

## Actual Behavior
- No caching of state
- Full query for every election
- 2-3 second election time
- Impact on availability

## Code Reference
- File: `server/election.go`
- Method: `electLeader()` - no state caching
- Missing: Local state cache
- Missing: Cache invalidation logic

## Additional Context
Level 2 performance issue. Fix requires:
- Implementing cluster state cache
- Cache invalidation on topology changes
- Incremental updates
- Performance benchmarking

## Suggested Labels
`performance`, `optimization`, `caching`, `leader-election`, `availability`

Start in server/election.go at electLeader() and trace how full cluster state is queried during leader election. Define the cache, topology-change invalidation, incremental updates, and background refresh from the issue requirements, then add performance benchmarks to verify election time falls below 100ms.

Written by the indexing model from the issue text.