kvcoord: optimize txn write buffer for read-only txns

[yuzefovich]

In the txn write buffer we need to merge the ScanResponse coming from the server with any overlapping writes that we've buffered. This requires us to allocate respMerger.batchResponses which is the "staging" area of the merged response. However, if we don't have any overlapping buffered writes, then we simply copy the server response into that staging area. This commit adds an optimization to avoid this extra copy in a special but common case when the buffer is empty (i.e. we're in a read-only txn). We could extend the optimization further but that is left as a TODO.

Epic: None
Release note: None

[blathers-crl]

It looks like your PR touches production code but doesn't add or edit any test code. Did you consider adding tests to your PR?

_{🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf.}

[cockroach-teamcity]

This change is 

[yuzefovich]

name                                                                     old time/op    new time/op    delta
ScanFilter/SharedProcessTenantCockroach/count1=25/count2=400/limit=50-8    1.71ms ± 2%    1.54ms ± 2%   -9.67%  (p=0.000 n=9+9)

name                                                                     old alloc/op   new alloc/op   delta
ScanFilter/SharedProcessTenantCockroach/count1=25/count2=400/limit=50-8     509kB ± 1%     329kB ± 2%  -35.24%  (p=0.000 n=9+9)

name                                                                     old allocs/op  new allocs/op  delta
ScanFilter/SharedProcessTenantCockroach/count1=25/count2=400/limit=50-8       912 ± 7%       748 ± 7%  -17.98%  (p=0.000 n=10+10)

[stevendanna]

Thanks for digging into this one. This seems like a pretty clear win.

[yuzefovich]

TFTR!

bors r+

[craig]

Build succeeded:

• examples_orms
• docker_image_amd64
• linux_amd64_fips_build
• acceptance
• unit_tests
• local_roachtest
• local_roachtest_fips
• check_generated_code
• lint
• linux_amd64_build