docs: clarify that text() max_nb_chars is an upper bound

[saimoon504]

Closes #2389

Brief summary: documentation-only fix. The text() docstring now states
explicitly that max_nb_chars is an upper bound, not a target.

What was wrong

While testing Faker for a university software testing course, we noticed
that text(max_nb_chars=200) regularly returns strings of 100-190
characters, and even max_nb_chars=99999 returned 99908 characters. We
reported this in #2389.

Reading the implementation, this is by design: the output is assembled
from whole words/sentences/paragraphs and the last unit that would
exceed the limit is dropped, so the result is always under max_nb_chars
but can be well under it. The docstring only said the length is
"approximate", which reads as "close to N" rather than "at most N,
possibly much less" - that is what confused us and what the issue is
really about.

How this fixes it

The docstring now states that:

• max_nb_chars is a strict upper bound
• the output can be substantially shorter, with an example
• values below 5 raise ValueError (already true in the code, but
  previously undocumented)

No behavior changes.