multithreaded gpu memory fill

• multithreaded gpu memory fill
• it is now possible to selec with -d card number larger than 9

MTP multi gpu version with only one scratch pad

This release follow up over the merging of the branch test into master.
There is a slight difference as this time, the several gpu loop are performed before checking for work update.

ccminer.exe is compiled with cuda 10 and compute_61,sm_61
ccminer_cuda8.exe is for older card (compiled with compute_52; sm_52)

experimental: multi GPU support with only one scratchpad (v2)

correct minor bug on linux side
add back cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync); which decrease cpu usage while mining (not during the initialization phase)

experimental: multi GPU support with only one scratchpad

Experimental, this release may crash. Give the possibility to run multiple GPU with only one scratch pad.
Maximum memory usage around 7Gb.

MTP multi GPU support

allows to run several gpu in one instance. however requires about 5Gb of ram per GPU

MTP release for mainnet (intensity adjustable)

intensity can be adjusted, the default is at 16 (ie 65536)
windows binaries compiled with cuda 10

MTP release for mainnet with less printout

windows binaries are for cuda 10 and may require a driver update

MTP release for mainnet

use correct funder address ready for mainnet (wallet and stratum)

new MTP release for testnet with no reinitilization

initialization occurs only once per job_id

stratum/wallet support win/linux

fix bugs in stratum/change endianness in gbt such it matches the one on stratum

linux stratum should work, although it hasn't been tested yet

fix timout issue when sending proof of work