tokenize: `e.offset` on `IndentationError` is not set correctly

Bug report

Bug description:

import io, tokenize

input_s = """
def foo():
    x = 1
   x = "some_very_long_expression"
"""
try:
    list(tokenize.generate_tokens(io.StringIO(input_s).readline))
except IndentationError as e:
    print(f"{e.offset = }")

The line x = "some_very_long_expression" is indented with 3 spaces instead of 4, so we get an error.
We would expect the offset to be around 3, since that is where the issue is.
On python 3.12+, the offset is instead 35, equal to the length of the line + 1.

This appears to be caused by the refactor of tokenize done in 3.12 for PEP 701
Potentially related to https://github.com/python/cpython/issues/84515
Funnily enough, the exact same thing seems to have happened around 3.8 https://github.com/python/cpython/issues/75676

Tested versions:

Testing command:
nix-shell -p python311 --run "python --version && python tmp.py"
(and replace 311 by 312, 313, etc)

CPython versions tested on:

3.15

Operating systems tested on:

Linux

Linked PRs

• gh-154281

Start with tokenize.generate_tokens and reproduce the supplied IndentationError example across the listed Python versions. Trace how the tokenizer constructs IndentationError and compare the offset with Python 3.11 behavior. Done means the misindented line reports an offset near 3, with regression coverage for the example.