check pledged src size at end of multithreaded frame

[avionicharshit-byte]

with nbWorkers >= 1 and an input spanning more than one job , a wrong pledged source size is never checked. the frame header carries the pledged size , compression returns success , and decompression fails later with Data corruption detected. both directions , too large and too small. the single-thread path has always caught this in ZSTD_compressEnd_public.

repro on current dev:

head -c 10000000 /dev/urandom > t.bin
cat t.bin | zstd -3 -T4 --stream-size=10000001 > out.zst ; echo $?   # 0
zstd -d out.zst -o /dev/null                                         # Data corruption detected
cat t.bin | zstd -3 --single-thread --stream-size=10000001 > /dev/null ; echo $?   # 11

#1095 set the rule that the pledge is validated whenever it is written into the frame header , and #1106 enforced it "whenever known" , but only in ZSTD_compressContinue_internal and ZSTD_compressEnd. the MT branch of ZSTD_compressStream2 was not part of it. this applies the same check there: at end of frame , compare pledgedSrcSizePlusOne with consumedSrcSize + 1 and return srcSize_wrong. both counters already exist , no new state.

tests: tests/cli-tests/compression/mt-stream-size.sh (over , under , exact , none with -T2 -B512K) and three cases in tests/zstreamtest.c with nbWorkers=2 and a 1 MB job size , streamed with ZSTD_e_continue so the first-call ZSTD_e_end override does not mask it. both fail on unfixed dev. make -C tests test passes.

one caveat: the error fires after the frame bytes have been handed to the caller , same as the single-thread path today.

[avionicharshit-byte]

@Cyan4973 small one: the MT path never validates pledgedSrcSize at end of frame , so -T4 --stream-size=<wrong> writes a bad frame and exits 0 while --single-thread returns 11. this adds the same check ZSTD_compressEnd has. is ZSTD_compressStream2 the right place for it , or would you rather have it in zstdmt_compress.c when the last job is created ?