fix(acl/openai): assign new block index for text interleaved between tool calls

[oierxjn]

Fixes #975

Problem

Some providers (observed on Cloudflare Workers AI gateways) emit content chunks between parallel tool-call chunks in one stream:

data: {"choices":[{"delta":{"tool_calls":[{"id":"call_a","index":0,"type":"function",...}]}}]}
data: {"choices":[{"delta":{"content":"\n"}}]}            <-- interleaved text
data: {"choices":[{"delta":{"tool_calls":[{"id":"call_b","index":1,"type":"function",...}]}}]}

chunkConverter.advanceContent only compares block type and source index, ignoring c.inToolCalls. Once tool calls have started, an interleaved text chunk still matches the previous text block's type, so it inherits the tool-call group's StreamingMeta.Index. schema.ConcatAgenticMessages then groups mismatched block types under one index and fails with:

content block type mismatch: expected 'function_tool_call', but got 'assistant_gen_text'

Fix

When inToolCalls is set, advanceContent now always advances to a new block index and resets inToolCalls, so:

• the interleaved text gets its own index (no collision with the tool-call group)
• a following tool call also gets a fresh index

Change is confined to (*chunkConverter).advanceContent; streams without interleaved text are unaffected.

Test

Added TestChunkConverter/text_interleaved_between_tool_calls covering the exact frame sequence (text → tool call 0 → interleaved text → tool call 1), asserting each block gets a distinct index and the full sequence passes ConcatAgenticMessages.

go test -gcflags="all=-N -l" ./libs/acl/openai/...   # ok

[CLAassistant]

All committers have signed the CLA.