Use read_all/pread_all where we expect full buffer to be read

[rvansa]

I've found that some GHA's in OpenJDK/crac are failing due to this error, and I think I've seen that even locally (on ia32) though rarely:

(00.040592) Error (criu/pagemap-cache.c:158): pagemap-cache: Can't read 9142's pagemap file: No such file or directory
(00.040595) Error (criu/pagemap-cache.c:173): pagemap-cache: Failed to fill cache for 9142 (ffe3b000-ffe3c000)
(00.040617) page-pipe: Killing page pipe
(00.040798) ----------------------------------------
(00.040804) Error (criu/mem.c:625): Can't dump page with parasite

Turns out that some parts of code use read(...) or pread(...) without handling that the call returns less bytes than the full size of the buffer; I've replaced those places where the expectation is obvious (error message if the returned value does not match buffer size) with read_all and new pread_all utility function.

[rvansa]

@wkia Please review.

[AntonKozlov]

This disturbs me, impact is not clear. Later we can process the code. Let's keep errno intact?

[rvansa]

This is actually a copy-paste of the read_all method above (just read -> pread). I don't see a problem in writing to errno?

[wkia]

@AntonKozlov
I agree with @rvansa - we could keep this consistent with other existing functions.
BTW I'd prefer using curly braces for one-line if's, but just a note.

[jankratochvil]

I would keep the @rvansa code. It is a copy-paste of read_all - while I would prefer its removal it would be a needless fork from upstream CRIU and it is not really required. If something then it could be rather an upstream CRIU contribution.
@wkia: CRIU does not use curly braces for one-line ifs: https://github.com/CRaC/criu/blob/crac/CONTRIBUTING.md#edit-the-source-code

[wkia]

@rvansa Sorry, missed this PR.

LGTM

[jankratochvil]

LGTM now.