### Issue

Large financial reports are slow to compute due to repeated re-aggregation of account_move_line balances for each formula, even when most formulas share the same base domain (usually filtered by account_id fields).

### Analysis

Each report line formula independently aggregates balances from account_move_line, even when their domains only differ on account_id-related fields such as account_id.account_type or account_id.non_trade.
This leads to redundant scanning and aggregation of the same dataset multiple times within a single report execution.

### Solution

Introduce a lightweight in-memory caching layer for aggregated balances by account_id, stored in self.env.cr.cache.

For each (options, date_scope) pair:
The report engine now computes once the mapping
{account_id: {'amt': total_balance, 'count_aml_lines': count}}.
This mapping is stored in the cursor cache and reused across all formulas whose domains filter exclusively on account_id fields.
A small domain transformation step allows AML domains based on account_id.* fields to be evaluated directly against the cached account aggregates.
This approach avoids redundant SQL aggregation, remains fully read-only (no database writes), and is safe for execution on read replicas.

### Benchmarks

Profiling get_report_information_readonly on different reports. Database has ~11.6 million account_move_lines, 305 account_accounts, and 16 account_types

| Report Name | Before | After | % Speed Up |
| --- |---|---|---|
| Balance Sheet | 35s | 6.2s | ~550% |
| Profit and Loss | 7.2 | 2.1 | ~300% |
| Cash Flow Statement | 1.5s | 0.4s | ~300% |
| Executive Summary | 24s | 6.3 | ~400% |

### References

opw-5130725