Expose precise cache-layer metrics for load_or_fetch

[blackopsrepl]

Problem

RoadNetwork::cache_stats() currently exposes hits and misses, but those counters do not correspond to the three cache layers described in the docs. In the current implementation:

• hits are only recorded for in-memory cache lookups
• misses are recorded before the file-cache check
• a single cold load can increment misses more than once
• a disk-cache hit is still counted as a miss

That makes the current counters unsuitable for operational guidance like "did this request hit memory, disk, or network?" or "how warm is the cache?"

Requested fix

Expose cache-layer metrics that match actual behavior, for example:

• load_requests
• memory_hits
• disk_hits
• network_fetches
• optionally in_flight_waits or similar contention metrics

These should let callers distinguish:

• served directly from the in-memory cache
• loaded from the on-disk cache
• fetched from Overpass

Why this matters

The docs site is adding operational guidance around caching. Without precise metrics, the docs either overclaim what cache_stats() means or have to stay vague. Tightening the crate API would make the docs accurate and make the metrics genuinely useful in production.

Current references

• src/routing/cache.rs
• src/routing/fetch.rs

In particular, the current behavior is visible around record_hit() / record_miss() in load_or_fetch, load_or_insert, and get_cached_network.

[blackopsrepl]

Closed by PR #27, merged into main at 2bfde95.