fix(parser): handle nested cache_creation.input_tokens format (v2.1.152+)

[delexw]

Summary

The Anthropic API can return prompt-cache write token counts in two formats:

• Flat (pre-v2.1.152): usage.cache_creation_input_tokens: N
• Nested (v2.1.152+): usage.cache_creation: { input_tokens: N }

Before this fix, all three token-scanning paths in the parser only read the flat field, so cache_creation_input_tokens was always 0 for sessions written by Claude Code v2.1.152 and later.

Changes

src-tauri/src/parser/entry.rs

• Add CacheCreationUsage struct for serde deserialization of the nested format
• Add optional cache_creation: Option<CacheCreationUsage> field to EntryUsage
• Add EntryUsage::effective_cache_creation_input_tokens() — returns max(flat, nested)
• Add cache_creation_from_value(usage: &Value) -> i64 helper for raw JSON Value access sites
• 7 new unit tests covering flat, nested, both, and neither formats

src-tauri/src/parser/classify.rs

• Use effective_cache_creation_input_tokens() instead of reading the flat field directly

src-tauri/src/parser/session.rs

• Replace inline .get("cache_creation_input_tokens") in both scan_session_metadata() and IncrementalTokenScanner::ingest_line() with cache_creation_from_value()
• 1 new integration test: incremental_scanner_nested_cache_creation_format

src-tauri/src/parser/subagent.rs

• Replace inline .get("cache_creation_input_tokens") in scan_subagent_tokens_into() with cache_creation_from_value()

specs/01-parser-pipeline.md

• Add v2.1.152 entry to the Version-Compatibility Normalisations table

Backward Compatibility

The max(flat, nested) strategy means:

• Old sessions (flat only, nested = 0) return the flat value unchanged
• New sessions (nested only, flat = 0) correctly return the nested value
• Both present returns the larger (safe edge case)

Fixes #116