Invalid UTF-8 values and ICU

Product: Tarantool
Since: 2.10.0-beta2
Audience/target: Tarantool users
Root document: https://www.tarantool.io/en/doc/latest/reference/reference_sql/sql_statements_and_clauses/#length
SME: @ ImeevMA

Details

[from commit message]:
The CHAR_LENGTH() and CHARACTER_LENGTH() functions now use ICU functions
to determine the length of a string.

[input from TarantoolBot]:
Invalid UTF-8 values may be handled differently depending on the ICU
version. For example, for this request:

SELECT CHAR_LENGTH(CAST(x'f0808080' AS STRING));

On centos 7 with libicu-devel-50.2-4.el7_7.x86_64 the result will
be:

tarantool> box.execute([[SELECT CHAR_LENGTH(CAST(x'f0808080' AS STRING));]])
---
- metadata:
  - name: COLUMN_1
    type: integer
  rows:
  - [1]
...

On ubuntu 20.04 with libicu-dev version 66.1-2ubuntu2 the result
will be:

tarantool> box.execute([[SELECT CHAR_LENGTH(CAST(x'f0808080' AS STRING));]])
---
- metadata:
  - name: COLUMN_1
    type: integer
  rows:
  - [4]
...

Requested by @ImeevMA in https://github.com/tarantool/tarantool/commit/c98e85319ba0603012cec974a838c1bc75cf6d0d.

Start with the CHAR_LENGTH section in the linked root document and compare it with the reported behavior for the invalid UTF-8 value x'f0808080'. Done means the documentation explains that results may vary by ICU version and includes the CentOS 7 and Ubuntu 20.04 examples shown in the issue.