1) quantification is needed for 2 purposes: to optimize reverse transcription and to check quality of RNA. For example less, than 50-100ng of RNA is not enough for RT. As for quality, most of spectrophotometers (including nanodrop) display RNA quality and asorbtion. Traces of proteins, salts and phenol inhibit RT and PCR, or may contain nucleases. 

2) hexamers are needed, when the target of interest is far from polyA tail, and you are not sure, that purified RNA is not degraded

3) Forward, as you have RNA (-) minus strand. But in fact you may use both primers. Once (+) is grown, the second (reverse) would increase the total concentration of cDNA region of interest.   

Let's assume you are talking i.e. about eukaryotic mRNA.  mRNA is, by definition,  considered to be sense in direction (+) because it exactly mimics/looks like the coding strand of the gDNA pieces that model it (although mRNA has U's instead of deoxy-T's of course). Since the opposite of the coding strand, called the gDNA "template strand" (anti-sense) strand is physically used as the "template" upon which mRNA/pre-mRNA is synthesized, the mRNA ends up being complimentary to that negative gDNA template strand; which gives the mRNA/pre-mRNA sense (+) orientation.  So I disagree with Andrey on that point. (This point gets missed by many).

So, since mRNA is single-stranded and sense (+) in orientation, if one wishes to use gene-specific priming, one can only use the reverse primer since the forward primer (which is sense (+)) is not complimentary to the sense (+) mRNA.  The reverse primer is called 'reverse" since it is negative (-) anti-sense in orientation; which is only complimentary to a sense (+) orientated target. The definition of "first-strand synthesis" is making single-stranded cDNA (which is anti-sense in direction) from single-stranded sense (+) mRNA by using the (-) reverse primer. Sometimes called the "down-stream' primer.

But, if instead priming with randomers and oligo dT:  a mixture of both is preferred in order to avoid 3' bias (by oligo dT priming only), and (as Andrey mentions above) to be sure that the 5' regions of the transcripts are represented robustly as well.

Also in agreement with Andrey, one should definitely measure RNA concentration before assembling RT reactions because of what Andrey mentioned above, and also because adding too much RNA to an RT reaction can kill the reaction (inhibit and/or entirely shut-down the reverse transcriptase enzyme; a well-known phenomena).  50 ng RNA/uL RT rxn is usually more than sufficient and not too much RNA to ensure good RT.  E.g., 1000 ng total RNA/20 uL RT = 50 ng/uL 

But, if the RNA has inhibitors in it, less RNA should be used. So, also in agreement with Andrey, ~200 ng total RNA/20 uL RT = 10 ng/uL is the lowest one should go in order to establish good RT.  The quality of the RT enzyme is extremely important here.  And, measuring the RNA integrity is important as well (e.g., measuring the RIN on a BioAnalyzer 2100 etc.) to prove that your RNA is intact. Spectrophotometer and NanoDrop will not tell you if the RNA is intact and/or undegraded. RIN values of at least 8.0 are desired - although RIN values as low as 5.5 can still work very well in qPCR as the small regions of interest are still largely intact as the law of averages apparently has it...

Thank you Dear Andery And Jack .