fix(solana): align daily balances merge grain

[0xRobin]

Fixes the incremental merge key for solana_utils.daily_balances so it matches the model's address/day output grain. Adds a dbt uniqueness test for that grain; existing duplicate rows will still require a full refresh/backfill of the table.

[cursor]

PR Summary

Medium Risk
Changes merge behavior on a core incremental balances table; incorrect historical rows may persist until a full refresh, affecting downstream models like latest balances and staking joins.

Overview
Aligns solana_utils_daily_balances incremental merge grain with how rows are actually produced — one snapshot per address per day (the model already picks the latest activity with row_number() partitioned by address and day, not by token mint).

The merge unique_address_key surrogate changes from address + token_mint_address + day to address + day, so merges update the same logical daily row instead of treating different mints as separate keys. A dbt_utils.unique_combination_of_columns test on day and address documents and guards that grain.

Note: Historical duplicate rows from the old key may still need a full refresh/backfill before the new test passes. A small YAML fix restores the leader column description indentation on solana_utils_block_leaders.

^{Reviewed by Cursor Bugbot for commit ce678a0. Configure here.}

[2bwfkjpr25-lab]

Thanks for addressing the incremental merge-grain mismatch!
This is only a contract clarification, not a request to expand this PR.
The current SQL produces one row per (address, day), and this PR aligns the merge key and uniqueness test with that implementation. However, the public Dune documentation describes each row as an end-of-day balance for a specific (address, token, day).
Could a maintainer confirm which grain is the intended public contract before this PR is finalized?

[2bwfkjpr25-lab]

Related follow-up: #9965 documents a separate state-semantics issue.

When the final balance-changing activity is lamport-only and its transaction metadata contains no token snapshot, coalesce(post_token_balance, 0) can store an unobserved token state as a known zero.

This is not a request to expand this PR. #9829 remains focused on aligning the incremental merge grain; historical duplicate rows still require the proposed full refresh/backfill.