fix: use catalog cache read multipliers in client savings

[Elioooon]

Summary

• generate each model's effective cache-read multiplier alongside client pricing
• use routed and requested model multipliers independently in Pi and Claude Code savings calculations
• include cache-read multipliers in the Pi pricing version and verify generated artifacts against the catalog

Fixes #1003.

Validation

• go test ./cmd/genprices ./internal/observability/otel
• bun test install/pi-router/test/savings.test.ts (8 passed)
• bash install/tests/cc-statusline_test.sh (47 passed)
• bash -n install/cc-statusline.sh install/install.sh
• git diff --check origin/main...HEAD

Scope

Cache creation remains at the existing 1.25x multiplier; this change is limited to cache reads. Missing prices or multipliers continue to fail closed instead of reporting fabricated savings.

[greptile-apps]

PR author is not in the allowed authors list.

[devin-ai-integration]

This conforms to our conventions — no changes needed, and thank you, this is a careful piece of work.

Reviewed against root AGENTS.md and the existing pricing patterns. A few things I liked: the multipliers come from catalog.Pricing.EffectiveCacheReadMultiplier() rather than a hand-maintained table, so internal/router/catalog stays the single source of truth; the cache_read block goes inside the generated BEGIN/END_GENERATED_PRICES region and is produced by cmd/genprices instead of being hand-edited; routed and requested sides each get their own multiplier (the asymmetry is the actual bug, and the pi + statusline tests pin it with a concrete number rather than recomputing the formula); the missing-multiplier case still fails closed to 0 savings instead of fabricating a number; and folding the multiplier into pricingVersion's canonical string means a multiplier-only catalog change still invalidates stale on-disk copies. Keeping cache creation at 1.25× and saying so in the header comment is the right scope call.

Verified locally: go test ./cmd/genprices ./internal/observability/otel pass, re-running the generator produces no drift from the artifacts committed here, the install.sh embedded copy is byte-identical to install/cc-statusline.sh's block, bash -n clean on both scripts, and install/tests/cc-statusline_test.sh is 47/47.