## Steps to reproduce the issue

1. Activate Mexican Localization
2. In Mexican Company, create two Invoices with one product line:
    - Price Unit of 3.47, 16% Tax
    - Activate CFDI to Public
3. Select both Invoices and create Global Invoice ("Actions" drop menu)
4. Check the CFDI tab in one of the Invoices
5. One of those errors pops up depending on your rounding method:

#### Round per line:
>     Code : 301
>     Message : Error de validaciones adicionales [Error #CFDI40108] El TipoDeComprobante es I,E o N, el importe registrado en el campo no es igual al redondeo de la suma de los importes de los conceptos registrados. Folio: 2. Serie: GINV/. El valor del atributo SubTotal (6.95) no coincide con la suma de los importes (3.47 + 3.47 = 6.94)

#### Round globally:
>     Code : CFDI40205
>     Message : El valor del campo TotalImpuestosTrasladados no es igual a la suma de los importes registrados en el elemento hijo Traslado.

## Explanation

#### Round per line:
With commit odoo/enterprise@933864a38af21a70b219d2824c64fff488acb15f, base and tax amounts were tweaked in order to satisfy some CFDI constraints. This change calculates amounts in an incorrect order: instead of calculating `tax_amount` with the formula `base_amount * tax_rate`, it calculates the base using `total / (1 + tax_rate)` then subtracts the base from the total to obtain the tax amount.
In our example, with a total of `6.94 + 1.12 = 8.06`, the base is calculated as such: `8.06 / 1.16 = 6.94827586207`, rounded to `6.95`, and the tax as such : `8.06 - 6.95 = 1.11`. The difference between the tweaked base `6.95` and the addition of the base of all invoice lines `3.47 + 3.47 = 6.94` is not accepted.

#### Round globally:
During a Global Invoice creation, we will first calculate the values of the Invoices separately. During this calculation, `total_impuestos_trasladados` is rounded with `precision_digits=2` while values in `traslados_list` are rounded with `precision_digits=6`. https://github.com/odoo/enterprise/blob/800378bfd3fe198647c9b418def3bd36066d9cd7/l10n_mx_edi/models/l10n_mx_edi_document.py#L583-L590 https://github.com/odoo/enterprise/blob/800378bfd3fe198647c9b418def3bd36066d9cd7/l10n_mx_edi/models/l10n_mx_edi_document.py#L950-L961 https://github.com/odoo/enterprise/blob/800378bfd3fe198647c9b418def3bd36066d9cd7/l10n_mx_edi/models/l10n_mx_edi_document.py#L963-L964 https://github.com/odoo/enterprise/blob/800378bfd3fe198647c9b418def3bd36066d9cd7/l10n_mx_edi/models/l10n_mx_edi_document.py#L1000-L1001
In the example, `3.47 * 0.16 = 0.552`, is rounded to `0.56`. The addition of every rounded value gives `1.12` or `1.110400`, which gives the difference between `total_traslados_impuestos` and `traslados_list`.

## Fix reasoning

#### Round per line:
We will add a hack in `_get_post_fix_tax_amount_map`: CFDI accepts a difference of `1e-<precision_digits>`, which means that `precision_digits=2` would allow a difference of `0.01`. If the difference is lower, we will not perform the operation and only round the values.
In our example, the obtained tax amount is `1.12` and the expected tax amount `6.94 * 0.16 = 1.1104`. The difference between the two is lower than `0.01` and is accepted by the CFDI.

#### Round globally:
We will use the calculated values in `_get_global_invoice_cfdi_values` to compute the total price and totals of taxes at the end of the method instead of using the rounded values received from `cfdi_values_list` at the start.

opw-4085403

[FIX] account_reports, l10n_*_reports: add explicit ORDER BY to report engines for consistent LIMIT and OFFSET behavior

As stated here https://www.postgresql.org/docs/current/queries-limit.html , postgresql requires an explicit ORDER by within the query for LIMIT and OFFSET to work properly and consistently between query executions. We hence add one to each engine not having one already.

=======================================

[FIX] account_reports: properly handle groupby in account_codes engine for grouping keys used with multiple accounts

// TO REPRODUCE

1) Create a report with a single line, whose computation uses the account_codes engine, with formula "1". Set "partner_id" as the groupby value for that line. Setup a "load more limit" of 2 on your report.

2) Create 3 accounts, with codes 11, 12, 13

3) Create 3 partners: A, B, and C

4) Create journal entries, with the following lines (all balancing lines must use other accounts as the ones defined in 2) ), in that order:
    - on account 11, with partner A, debit=10
    - on account 11, with partner A, debit=20
    - on account 12, with partner A, debit=25
    - on account 11, without partner, debit=30
    - on account 11, without partner, debit=40
    - on account 12, with partner B, debit=50
    - on account 13, with partner C, debit=60

5) Open the report for a period covering all the created journal entries, and have a look at the sublines generated for the groupby.

    => With the limit being 2, you'd expect to see "A", "B", and a "Load more..." line allowing to visualize the rest (namely a line for "C", and "Unknown"). Instead of that, you see no "Load more ...", and line "A"'s amount is wrong.

// EXPLANATION

The SQL query of the account_codes engine is grouping not only by groupby key (here, partner_id), but also by account_id, to be able to process the "C" and "D" suffixes of the formula later on. Because of that, the LIMIT passed by the load_more_limit is applied on that grouping.

With a load_more_limit of 2, we try to load 3 lines. If we manage to load 3, we only display 2, but know we need to display the "Load more..." line. Else, we know there are no more results, so no need for that additional line.

So, with our example, the query runs with a LIMIT of 3, returning
    - 30 for A on account 11
    - 25 for A on account 12
    - 70 for no partner on account 11

Then, another issue happens:

The results returned by the SQL query (hence, by the engine) contain two times the same partner_id. From an engine perspective, the grouping key is only partner_id, not account_id (which is only used for an inner computation, with the SQL query). So that means the same grouping key has two different values. This is not legit, and causes one of them to shadow the other.

The report computation then receives 1 result for A, and 1 for "Unknown", so it renders them, and since there is no third result, it does not display the "Load more ...".

Forward-Port-Of: odoo/enterprise#74784