fix: Raise on data mismatch in str.json_decode

[nameexhaustion]

Fixes #13061

• Raise if JSON data cannot be decoded to dtype instead of silently appending NULL
• Raise if there are extra keys in struct data outside of the schema instead of silently ignoring them

print(pl.Series([None, "1"]).str.json_decode(infer_schema_length=1))
# Before
# shape: (2,)
# Series: '' [null]
# [
# 	null
# 	null
# ]
#
# After
# polars.exceptions.ComputeError: error deserializing JSON: error deserializing value "Static(U64(1))" as null. \
#             Try increasing `infer_schema_length` or specifying a schema.
print(
    pl.Series(['{"a": 1}', '{"a": 2, "b": 2}']).str.json_decode(infer_schema_length=1)
)
# Before
# shape: (2,)
# Series: '' [struct[1]]
# [
# 	{1}
# 	{2}
# ]
#
# After
# polars.exceptions.ComputeError: error deserializing JSON: extra key in struct data: b

Todo: I think our NDJSON and JSON readers also suffer from this issue

[codecov]

Codecov Report

Attention: Patch coverage is 80.69307% with 39 lines in your changes missing coverage. Please review.

Project coverage is 80.18%. Comparing base (11bd08e) to head (8f4d95a).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/polars-json/src/json/deserialize.rs	82.94%	29 Missing ⚠️
crates/polars-json/src/ndjson/deserialize.rs	56.52%	10 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #19347      +/-   ##
==========================================
- Coverage   80.21%   80.18%   -0.03%     
==========================================
  Files        1523     1523              
  Lines      209979   210096     +117     
  Branches     2432     2432              
==========================================
+ Hits       168426   168472      +46     
- Misses      40997    41068      +71     
  Partials      556      556              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[nameexhaustion]

We currently have this behavior where a user can perform a column selection within a struct field by giving a partial schema - this happens because of the existing behavior of JSON deserialization to silently ignore extra fields.

In this PR we change that behavior to raise an error, but to maintain the ability to do this column selection I made it so that we don't raise when we see extra keys if the schema was provided by the user.

[nameexhaustion]

Ensure we only push NULL if the value itself was "null", and add tracking of an "error index" to all of the deserializers that tells us the position of the first parsing error.