fix(streaming): enforce strict index validation in event accumulator to prevent state corruption

[l1ve709]

Summary

This PR resolves an unhandled edge case in the streaming parser where out-of-sequence or dropped content_block_start events could silently corrupt the internal accumulator state, leading to cryptic IndexError exceptions during downstream content_block_delta processing.

Root Cause Analysis

The accumulator previously appended content_block_start payloads blindly without asserting the sequence invariant (as noted by the legacy # TODO: check index comment). If an SSE event is dropped or misordered by the network layer, current_snapshot.content desynchronizes from the expected block sequence.

When a subsequent content_block_delta arrives, the parser relies on content = current_snapshot.content[event.index]. If the internal state length is misaligned with the incoming event.index, it triggers an obscure IndexError inside the delta router, making it extremely difficult for developers to debug the underlying network/API failure.

Changes Made

• Introduced a strict sequence invariant check during content_block_start evaluation for both standard (_messages.py) and beta (_beta_messages.py) streams.
• The parser now employs a fail-fast mechanism by explicitly raising a RuntimeError describing the index mismatch, guaranteeing stream integrity.

Validation

• Adheres to existing ruff static analysis and formatting guidelines.
• The state machine mathematically ensures event.index == len(current_snapshot.content) before allowing .append() mutations.