Set cache breakpoint on second-to-last message (#27632)

Here's a sample `dbg!` of token usage after this change, for a small
agent thread:

```
[crates/assistant2/src/thread.rs:1092:25] &usage = TokenUsage {
    input_tokens: 5354,
    output_tokens: 184,
    cache_creation_input_tokens: 0,
    cache_read_input_tokens: 0,
}
[crates/assistant2/src/thread.rs:1092:25] &usage = TokenUsage {
    input_tokens: 54,
    output_tokens: 132,
    cache_creation_input_tokens: 5518,
    cache_read_input_tokens: 0,
}
[crates/assistant2/src/thread.rs:1092:25] &usage = TokenUsage {
    input_tokens: 54,
    output_tokens: 113,
    cache_creation_input_tokens: 166,
    cache_read_input_tokens: 5518,
}
[crates/assistant2/src/thread.rs:1092:25] &usage = TokenUsage {
    input_tokens: 291,
    output_tokens: 181,
    cache_creation_input_tokens: 147,
    cache_read_input_tokens: 5684,
}
```

Release Notes:

- N/A
edc7d73643Richard Feldman committed on 3/27/2025, 10:32:50 PM· committed by GitHubparent4839195
1 file changedLine totals unavailable