feat: add new context pricing tiers

[rekram1-node]

Summary

Adds authored context pricing tiers so model configs can express long-context pricing at provider-specific thresholds instead of being tied to the legacy 200k field.

New Authored Format

Base pricing remains on [cost] and continues to mean USD per 1M tokens:

[cost]
input = 5
output = 25
cache_read = 0.5

Additional context-based pricing is authored with [[cost.tiers]]:

[[cost.tiers]]
context = { min = 272_000 }
input = 10
output = 45
cache_read = 1

The context.min value is the request context threshold where that price applies. This allows thresholds like 272_000, 256_000, 200_000, or provider-specific values without encoding them in the field name.

Compatibility

• Authored TOML should use [[cost.tiers]] for long-context pricing.
• Authored TOML rejects the old [cost.context_over_200k] field.
• Generated/API output still includes cost.context_over_200k when a single compatible open-ended context tier can be represented for legacy clients.
• The generated output also preserves the new cost.tiers data for clients that want the exact threshold.

Notes

• Existing base cost.input, cost.output, and cache price fields are unchanged.
• Invocation variants such as fast/slow modes remain under experimental.modes.*; this PR only models automatic context-based price changes within a cost schedule.