Check columns when creating new chunk from table

[melihmutlu]

Chunks must have the same columns with their hypertables. create_chunk()
should not allow to have a chunk with additional columns which do not
exist in the parent hypertable. Check columns by name in create_chunk()
when creating a chunk from an existing table.

This commit also forces chunks and hypertable to have the same
expressions for generated columns, even though PostgreSQL inheritance
may allow it.

For PG15, it is allowed to have generated column in chunks if and only
if the corresponding hypertable column is also generated. For later
versions, this is handled by PostgreSQL code.

[codecov]

Codecov Report

Attention: Patch coverage is 75.86207% with 7 lines in your changes missing coverage. Please review.

Project coverage is 82.41%. Comparing base (bc69773) to head (276b7d6).
Report is 100 commits behind head on main.

Files with missing lines	Patch %	Lines
src/chunk.c	76.19%	0 Missing and 5 partials ⚠️
src/utils.c	75.00%	0 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #8164      +/-   ##
==========================================
+ Coverage   82.32%   82.41%   +0.09%     
==========================================
  Files         256      256              
  Lines       47990    47988       -2     
  Branches    12095    12102       +7     
==========================================
+ Hits        39506    39551      +45     
- Misses       3635     3646      +11     
+ Partials     4849     4791      -58     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[erimatnor]

Just minor nits and suggestions. Otherwise LGTM.

[mkindahl]

I agree with the other reviewers that the locks are too strong and you should just use locks strong enough to prevent modifications of the table definition.

I am also missing tests for the following (some where missing from before, but it's good to have checks for them any way). I might have missed that some of them exists, already, so feel free to point me to them in that case.

• Create a chunk but use a table that you created with more columns than necessary and then altered to drop some columns. This is to catch cases where the attribute numbers do not match but the names match. (You are checking attisdropped, so it should be fine, but good to test it.)
• Create a chunk that has all the right names and types, but in a different order than the parent.
• Create a table with too few columns, insert some data, and add a new column with the correct type to match the parent hypertable and then try to add it. This is to see what happens with attributes that has atthasmissing set.
• Add a chunk that has all the correct names compared to the parent table, but with different types. I don't see any such tests in the current test, so might be good to add it since you're making this change.
• Add a chunk with a generated column of the correct name and type.

After you've added chunks that should be possible to add, you should probably try to insert some rows in the parent table and see that they are routed correctly (granted, these were missing since before).

[melihmutlu]

See the commit postgres/postgres@8bf6ec3 for the changes related to restrictions on generated columns in inheritance after PG15.

[mkindahl]

A few suggestions for extending the tests and a few nits. Otherwise looks good.