fix: avoid panics on Jinja-templated SQL with config()/source() blocks

[kimfrie]

Summary

• Fixes the Option::unwrap() on a None value panic at crates/lib/src/utils/reflow/elements.rs:272 that occurs when every newline in a ReflowPoint comes from a Jinja template block (e.g. the source combines {{ config(...) }} with {{ ref() }} / {{ source() }} on a referenced table).
• When no literal newline exists in the point, indent_to now emits no fix and returns the point unchanged, rather than crashing the linter run.
• Also hardens the inner get_position_marker().unwrap() used while matching the literal newline.

Fixes #2464.

Reproduction

Using the minimal example from #2464 (dbt templater, bigquery dialect):

{{ config(materialized='table') }}

with raw_data as (
    select * from {{ source('conn', 'tbl') }}
               )
select *
from raw_data

Before: thread '<unnamed>' panicked at crates/lib/src/utils/reflow/elements.rs:272:22: called Option::unwrap() on a None value.
After: linter completes with ordinary LT-rule violations, no panic.

Verified locally against 12 real-world dbt/jinja models (redshift dialect, apply_dbt_builtins = True) that each reproduced the panic on sqruff 0.37.3 — all now lint cleanly.

Test plan

• cargo test --release --features python -p sqruff-lib --lib reflow — all 10 reflow tests pass
• Manually re-linted 12 files from a downstream project; 0 panics vs. dozens previously

🤖 Generated with Claude Code

[kimfrie]

@benfdking would you help me with a review?