Jinja - add missing features for DeepSeek R1

[giladgd]

This PR adds support for the following Jinja features:

• Accessing members of call expression results
• .split on strings - the implementation matches Python's str.split(sep=None, maxsplit=-1) function behavior

DeepSeek R1's chat template used to fail on this expression, which now works in this PR:

{%- set content = content.split('</think>')[-1] -%}

Fixes #1141

[giladgd]

I'm not sure why the CI tests don't run on my branch, but they pass when I run them on my machine

[xenova]

Hey! I'm in the process of reviewing the PR and I'll trigger the CI shortly!

I found a few minor issues with the current implementation, so I'm just fixing a few things.

[xenova]

Thanks for the PR! I cleaned up & fixed a few issues with the implementation of split, but other than that, it looks good!

If you're interested, here were the failing strings, which I've now added as unit tests

SPLIT_4: `|{{ "  1 2  3   ".split() | tojson }}|{{ "babbaccabbb".split("b") | tojson }}|{{ "babbaccabbb".split("b", 2) | tojson }}|`,
SPLIT_5: `|{{ " 1 2 3 4 5 ".split(none, 0) | join(",") }}|{{ " 1 2 3 4 5 ".split(none, 3) | join(",") }}|{{ " 1 2 3 4 5 ".split(" ", 0) | join(",") }}|{{ " 1 2 3 4 5 ".split(" ", 3) | join(",") }}|{{ " 1 2 3 4 5 ".split(" ", 10) | join(",") }}|`,

[xenova]

I'll merge after a final check by @giladgd :)

[giladgd]

@xenova LGTM
Thanks for implementing join BTW, it was the next thing I was planning to do :)

[xenova]

I just updated the e2e unit test to include text so that the split branch is taken. Ready to merge now!

Thanks again!