fix(harness): apply an edited max_turns to the scenario a run plays

[KarthikAvinashFI]

Why

A run lays the platform's current scenario edits over the sealed scenario document before playing it (_with_scenario_edits in fi/alk/harness/call_runner.py). The overlay only takes the fields listed in _EDITABLE_FIELDS, and max_turns was not listed. So an edited turn budget reached the run's metadata but never the chat call that reads it.

What changed

File	Change
src/fi/alk/harness/call_runner.py	_EDITABLE_FIELDS includes max_turns. Voice calls do not read max_turns, so they are unaffected; chat calls (chat_call_runner._conversation_scenario) now use the edited value.
tests/harness/test_scenario_edits.py	The overlay test and the chat caller test assert the edited max_turns, plus a persona languages list with two entries.

Tests

Test	Scenario it pins
test_edits_replace_only_what_a_call_plays	max_turns and a two-language persona are overlaid; tests and instruction stay sealed.
test_a_chat_caller_is_the_edited_persona	The chat scenario built from the merged document carries the edited max_turns.

Verified live

Built into a hosted sandbox template with ALK dev and #128. The image carries _EDITABLE_FIELDS = ("background_noise", "max_turns"). An edited max_turns reaches the run's scenario_edits on a chat environment. Voice calls, which do not read max_turns, completed on the new template across two fresh environments.

How to run

pytest tests/harness/test_scenario_edits.py tests/harness/test_call_runner.py tests/harness/test_chat_call_runner.py

Linear

Part of TH-7985

[KarthikAvinashFI]

Folded into #128.