repository: add obj_offset/obj_size range-load params to Store.load calls

[mr-raj12]

Description

get() now accepts obj_offset=0 and obj_size=None and forwards them to all store.load() calls. The values come from PackWriter's flush results; once update_pack_info() has written them into the chunk index, callers can fetch a single chunk out of a multi-chunk pack without loading the whole file.

When obj_size is set, the read_data=False path caps the initial 1 KB load to obj_size so the store is not asked for more bytes than the object contains. The retry load is also capped as a guard against a corrupted meta_size in the header; for a well-formed object the clamp is a no-op.

get_many() gets a TODO comment: once N>1 packs land, it and fetch_many will need per-id (pack_id, obj_offset, obj_size) tuples to range-load each chunk.

Changes:

• repository.py: get() takes obj_offset=0, obj_size=None; both read_data branches pass offset=/size= to store.load().
• testsuite/repository_test.py: test_get_with_range writes a two-chunk pack via store_store, then range-loads each chunk by offset and size.

refs #8572

Checklist

• PR is against master
• New code has tests and docs where appropriate
• Tests pass
• Commit messages are clean and reference related issues

[codecov]

Codecov Report

❌ Patch coverage is 40.00000% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.72%. Comparing base (35eff1c) to head (e4b4f47).
⚠️ Report is 4 commits behind head on master.

Files with missing lines	Patch %	Lines
src/borg/repository.py	40.00%	5 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #9744      +/-   ##
==========================================
- Coverage   84.76%   84.72%   -0.05%     
==========================================
  Files          92       92              
  Lines       15047    15053       +6     
  Branches     2250     2252       +2     
==========================================
- Hits        12755    12754       -1     
- Misses       1592     1597       +5     
- Partials      700      702       +2     

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

[ThomasWaldmann]

From the PR comment:

The read_data=False path clamps both load sizes to obj_size when set. Right now with N=1 packs this changes nothing (one chunk per file), but once N>1 packs land an unclamped size would overshoot into the next chunk,
so the clamp goes in now.

This is rather confusing.

For the read_data=False path, it first reads 1KB (assuming that this usually contains the header and all the metadata), an overshoot into next object is no problem as the parse_meta function will only read the metadata using the correct length from the header and will ignore the trailing bytes.

If meta_size in the header tells that we did not read enough data, we do a 2nd attempt, this time with exactly to correct size for what we need.

The whole point of doing it like this is to avoid just reading the header (few bytes) and then having to do another read for just the few bytes of metadata. 2x latency.